Coastal city underground space resource development suitability assessment method based on multi-source data driving

Abstract

The application relates to a coastal city underground space resource development suitability evaluation method based on multi-source data driving, which comprises the following steps: determining a target region and establishing multi-dimensional evaluation indexes; obtaining corresponding multi-source data sets and performing space-time alignment processing to generate standardized data sets; determining the first-level evaluation weights of each first-level index in the multi-dimensional evaluation indexes based on the standardized data sets; determining and optimizing the grading thresholds of each first-level index in the multi-dimensional evaluation indexes based on the standardized data sets; performing weighted fusion calculation according to the first-level evaluation weights and the grading thresholds to generate underground space development scores; determining the suitability grades based on the underground space development scores and outputting engineering suggestion information; and in summary, the application integrates the first dynamic evaluation factor and the second dynamic evaluation factor, performs combined weighting and grading threshold optimization based on the standardized data sets, effectively processes dynamic environmental factors such as typhoons, and thus improves the reliability of the evaluation.

Description

Technical Field

[0001] This application relates to the technical field of resource development assessment, and in particular to a method for assessing the suitability of underground space resource development in coastal cities based on multi-source data. Background Technology

[0002] In the field of underground space development and safe utilization, suitability assessment for underground space resource development, as a preliminary step in engineering planning, plays a decisive role in ensuring construction safety, optimizing spatial layout, and mitigating potential risks. Against this backdrop, the complex and dynamic geological environment of coastal cities places increasingly higher demands on the real-time data processing capabilities and multi-source heterogeneous data fusion technology of assessment systems.

[0003] Existing suitability assessment systems for underground space resource development generally adopt multi-dimensional indicator systems, integrating static parameters such as geological conditions, environmental factors, and development needs, and completing the assessment through conventional data processing methods, such as the analytic hierarchy process, fuzzy comprehensive evaluation method, or GIS spatial overlay analysis. However, such methods have gradually revealed their shortcomings when dealing with the complex dynamic environment of coastal cities, especially against the backdrop of frequent extreme weather events such as typhoons and storm surges. Specifically, traditional resource development assessment systems rely excessively on historical average data and fixed thresholds, failing to consider the chain reactions of sudden changes in groundwater levels and intensified soil erosion caused by typhoons. This leads to a severe underestimation of dynamic risk factors. For example, the combined effects of heavy rainfall and storm surges can significantly alter hydrogeological conditions, inducing ground subsidence or piping. However, the existing system's indicator framework lacks a real-time quantification mechanism for typhoon impact parameters, causing the assessment results to be disconnected from actual engineering risks. Furthermore, the setting of key parameters such as indicator weights and grading thresholds in existing technologies is often limited to single algorithm outputs or expert subjective judgments, failing to incorporate data-driven dynamic verification and iterative optimization based on engineering cases. In summary, this existing static processing method is ill-suited to the rapid evolution of geological conditions during special scenarios such as typhoon seasons. This results in significant deficiencies in the assessment system's technical aspects, such as real-time dynamic data processing, multi-source heterogeneous data fusion, and adaptive optimization of key parameters. Consequently, it struggles to meet the engineering and technical requirements of underground space development in coastal cities, thereby increasing safety hazards in underground space development. Summary of the Invention

[0004] To address the aforementioned shortcomings, this application provides a method for assessing the suitability of underground space resource development in coastal cities based on multi-source data.

[0005] The above-mentioned objective of this application is achieved through the following technical solution: A method for suitability assessment of underground space resource development in coastal cities based on multi-source data, comprising the following steps: The target area is identified and a multidimensional evaluation index is established, which includes several primary indicators. The primary indicators include engineering geological conditions, hydrogeological conditions, adverse geological indicators, and urban development demand indicators. The hydrogeological conditions and adverse geological indicators respectively integrate a first dynamic evaluation factor and a second dynamic evaluation factor. The system acquires multi-source datasets of the target area in real time and performs spatiotemporal alignment processing on the multi-source datasets to generate a standardized dataset. The standardized dataset includes geological exploration data, meteorological monitoring data, urban status data, and planning and management data. Based on a standardized dataset, the first-level evaluation weights of each first-level indicator in the multidimensional evaluation index are determined by a pre-set combination weighting strategy, and the grading thresholds of each first-level indicator in the multidimensional evaluation index are determined by a pre-trained grading threshold optimization model. The standardized dataset is weighted and fused according to the primary evaluation weight and the hierarchical threshold to generate an underground space development score for the target area. The suitability level of the target area is determined based on the underground space development score, and a development decision assessment report corresponding to the suitability level is generated.

[0006] In summary, the proposed method for assessing the suitability of underground space resource development in coastal cities based on multi-source data integration can effectively capture geological changes under extreme events such as typhoons by integrating dynamic assessment factors, real-time data acquisition, and weighted fusion calculation. This addresses the shortcomings of existing static assessment systems in dynamic response and data integration, and improves the accuracy and real-time nature of assessments, while reducing the risks and hidden dangers of resource development under extreme events such as typhoon seasons. Attached Figure Description

[0007] Figure 1 This is a flowchart of an embodiment of a method for assessing the suitability of underground space resource development in coastal cities based on multi-source data driven by this application; Figure 2 This is a flowchart of step S10 in an embodiment of a method for assessing the suitability of underground space resource development in coastal cities based on multi-source data in this application. Figure 3 This is a flowchart of step S13 in an embodiment of a method for assessing the suitability of underground space resource development in coastal cities based on multi-source data driven by this application. Detailed Implementation

[0008] In existing underground space resource development suitability assessment systems, the over-reliance on historical average data and fixed thresholds fails to effectively integrate the dynamic geological effects caused by extreme weather events such as typhoons. This leads to a serious underestimation of cascading risk factors such as sudden changes in groundwater levels and intensified soil erosion. Furthermore, the lack of real-time quantification of typhoon impact parameters in hydrogeological conditions and adverse geological indicators easily results in a disconnect between assessment results and actual engineering risks. At the same time, the setting of key parameters is limited to single algorithm outputs or expert subjective judgments, without data-driven dynamic verification and iterative optimization based on engineering cases. Consequently, existing assessment systems suffer from technical deficiencies in real-time dynamic data processing, multi-source heterogeneous data fusion, and parameter adaptive optimization, thus affecting the accuracy of assessment results.

[0009] For example, during typhoon season in coastal cities, when the combined effects of heavy rainfall and storm surges cause rapid changes in hydrogeological conditions, existing systems, lacking a dynamic screening mechanism for typhoon disturbance parameters, cannot promptly capture abnormal fluctuations in groundwater levels and dynamic changes in soil erosion. This leads to the neglect of the correlation between engineering geological conditions and hydrogeological conditions, resulting in deviations in suitability level assessments and affecting the scientific validity and safety of underground space development decisions. Furthermore, meteorological monitoring data and geological exploration data from multi-source datasets are prone to misalignment due to inconsistencies in spatiotemporal benchmarks, causing a loss of key information during the standardization process and rendering the grading thresholds unable to adapt to the rapid evolution of geological conditions.

[0010] If the above problems are not addressed, the reliability of the underground space resource development suitability assessment system will continue to decline, potentially leading to the neglect of potential risks during underground space development and increasing the likelihood of construction safety accidents.

[0011] In this regard, this application discloses a method for assessing the suitability of underground space resource development in coastal cities based on multi-source data. In one embodiment, such as... Figure 1 As shown, the specific steps include the following: S10: Determine the target area and establish a multi-dimensional evaluation index including several primary indicators. The primary indicators include engineering geological conditions, hydrogeological conditions, adverse geological indicators, and urban development demand indicators. The hydrogeological conditions and adverse geological indicators respectively integrate a first dynamic evaluation factor and a second dynamic evaluation factor. In this embodiment, the target area refers to a specific coastal city or region requiring an assessment of the suitability of underground space resource development. This target area has clearly defined geographical boundaries and is susceptible to extreme weather events such as typhoons. Multidimensional assessment indicators refer to evaluation standards used to measure the suitability of underground space development across multiple dimensions. In this embodiment, the multidimensional assessment indicators are hierarchical, such as primary and secondary indicators, to comprehensively reflect various factors affecting suitability. Primary indicators are the highest-level indicators in the multidimensional assessment indicator system. These primary indicators include engineering geological conditions, hydrogeological conditions, adverse geological indicators, and urban development needs indicators. Engineering geological conditions refer to the physical and mechanical properties of the site's soil and rock mass, which are the basis for assessing the stability of the engineering foundation and the ease of excavation. Secondary indicators may include the natural unit weight of the soil and rock mass, characteristic values ​​of the foundation bearing capacity, rock mass integrity index, and soil moisture content. Hydrogeological conditions refer to geological environmental conditions related to groundwater, including groundwater... The evaluation criteria include: the occurrence and movement patterns of groundwater and their impact on engineering projects; secondary indicators such as groundwater static water level and corrosivity level; and integration of the first dynamic evaluation factor. Adverse geological indicators refer to geological phenomena that may trigger geological disasters or adversely affect engineering construction; secondary indicators such as fault density, sand liquefaction discrimination index, and soft soil distribution thickness; and integration of the second dynamic evaluation factor. Urban development demand indicators refer to the requirements and constraints imposed on underground space development by urban planning and socio-economic factors; secondary indicators such as distance from rail transit stations, benchmark land price, population density, and planned land use nature; ensuring that the evaluation results are aligned with urban development goals. The first and second dynamic evaluation factors refer to parameters that can cause significant changes in hydrogeological conditions or adverse geological indicators under specific external conditions. The first and second dynamic evaluation factors are integrated into the corresponding primary indicators to reflect the response of the evaluation object in a dynamic environment.

[0012] Specifically, in practical applications, the target area can be determined based on the boundary delineation of a geographic information system, for example, by inputting the latitude and longitude range or selecting a zoning area. The establishment of multidimensional evaluation indicators can be initially set by expert experience and relevant requirements. For example, engineering geological conditions can include stratum lithology and bearing capacity, hydrogeological conditions can include groundwater level and permeability coefficient, adverse geological indicators can include soft soil distribution and fault zones, and urban development demand indicators can include population density and transportation demand. Among them, hydrogeological conditions and adverse geological indicators respectively integrate the first dynamic evaluation factor and the second dynamic evaluation factor. The first dynamic evaluation factor and the second dynamic evaluation factor can be initially identified and integrated based on historical observation data through statistical analysis methods to reflect their potential impact on the corresponding primary indicators.

[0013] S20: Real-time acquisition of multi-source datasets of the target area, and spatiotemporal alignment processing of the multi-source datasets to generate a standardized dataset, which includes geological exploration data, meteorological monitoring data, urban status data and planning management data; In this embodiment, a multi-source dataset refers to a collection of raw data obtained from multiple different channels or systems. The various data in a multi-source dataset may have different formats, structures, timestamps, and spatial references, requiring appropriate preprocessing before evaluation. Spatiotemporal alignment refers to the unification of data from different sources in terms of time and space dimensions to ensure comparability and consistency across all data. Spatiotemporal alignment includes operations such as coordinate transformation, time series normalization, and data interpolation. A standardized dataset refers to a collection of data that has undergone spatiotemporal alignment, outlier correction, and format unification. Standardized datasets include geological exploration data, meteorological monitoring data, urban status data, and planning management data. Geological exploration data refers to raw data and results obtained through engineering geological exploration methods, used to reveal the structure, composition, and physical and mechanical properties of underground soil and rock masses in the evaluation area. This mainly includes borehole columnar sections and soil and rock mass tests. Reports and 3D geological models are the most direct basis for assessing the stability of the engineering foundation and the ease of excavation. Meteorological monitoring data refers to observational data obtained by meteorological, hydrological and other relevant departments, reflecting the climate conditions of the assessment area, especially extreme weather events. It mainly includes key characteristic parameters such as typhoon path, intensity, and central pressure, as well as long-term groundwater monitoring data. Urban status data refers to objective data describing the current social development and spatial utilization of the target area, mainly including population density, benchmark land price, and the distribution of existing infrastructure. It reflects the existing functional layout and development intensity of the city and is key to assessing development needs and economic benefits. Planning management data refers to spatial control and land use data formulated by planning-related departments to guide and constrain the future development of the city. It mainly includes urban master plan texts, ecological protection restriction vector data, and planned land use nature. It defines the boundaries and directions of development and ensures that the assessment results meet the city's long-term development goals and protection requirements.

[0014] Specifically, multi-source datasets can be acquired in various ways. For example, geological exploration data can be exported from existing databases; meteorological monitoring data can be obtained from regularly published reports from meteorological stations; urban status data can be obtained from relevant urban management departments; and planning management data can be obtained from publicly available documents from these departments. When performing spatiotemporal alignment, data from different sources can be matched based on their recorded time and location. For the time dimension, a unified time granularity can be set, such as monthly, and all data can be consolidated to that granularity. For the spatial dimension, data from different coordinate systems can be converted to a pre-defined unified coordinate system and spatially matched. Standardized datasets can be generated by converting the aligned data into a format and unifying the units, for example, converting all data into tabular form and standardizing the units.

[0015] S30: Based on a standardized dataset, the first-level evaluation weights of each first-level indicator in the multidimensional evaluation index are determined through a pre-set combined weighting strategy, and the grading thresholds of each first-level indicator in the multidimensional evaluation index are determined through a pre-trained grading threshold optimization model. In this embodiment, the combined weighting strategy refers to a pre-defined comprehensive method for determining the weights of evaluation indicators. This method may combine multiple weighting techniques, such as subjective weighting and objective weighting, to obtain more reasonable and accurate indicator weights. The first-level evaluation weight refers to the relative importance of each first-level indicator in the multi-dimensional evaluation indicator system, used to balance the influence of different first-level indicators in weighted fusion calculations. The hierarchical threshold optimization model refers to a trained machine learning model that can determine and optimize the level classification thresholds of each evaluation indicator based on historical data and current conditions, aiming to improve the accuracy and adaptability of the evaluation results. The hierarchical threshold refers to the critical value that divides the value range of the evaluation indicator into different suitability levels, and is a key parameter for determining the suitability level.

[0016] Specifically, the combined weighting strategy can employ a combination of the analytic hierarchy process (AHP) and expert scoring. For example, a judgment matrix can be constructed first through expert scoring, and then the subjective weights of each primary indicator can be calculated. The determination of the grading thresholds can be based on industry standards or rules of thumb. For instance, local requirements for underground space development can be referenced to directly set the grading thresholds for each indicator. Alternatively, the suitability grading boundaries for each indicator can be determined based on the experience of historical engineering cases and expert advice. The pre-trained grading threshold optimization model can be a linear regression model, which obtains a preliminary threshold adjustment function by fitting historical data.

[0017] S40: Based on the primary evaluation weights and hierarchical thresholds, the standardized dataset is weighted and fused to generate an underground space development score for the target area. In this embodiment, weighted fusion calculation refers to combining the evaluation results of multiple indicators into an overall evaluation score by weighted summation or other aggregation methods based on the weights and scores of each indicator; the underground space development score is a quantitative score that reflects the suitability of underground space development in the target area, obtained through weighted fusion calculation. The higher the score, the better the suitability.

[0018] Specifically, the standardized dataset is weighted and fused according to the primary assessment weights and grading thresholds to generate an underground space development score for the target area. The weighted fusion calculation can be performed using a weighted summation method, that is, the standardized value of each primary indicator is multiplied by its corresponding primary assessment weight, and then the weighted results of all primary indicators are summed to obtain the final underground space development score.

[0019] S50: Determine the suitability level of the target area based on the underground space development score, and generate a development decision assessment report corresponding to the suitability level.

[0020] In this embodiment, suitability level refers to different development suitability categories divided according to the underground space development score and a preset grading threshold. For example, it can be divided into "high suitability", "medium suitability", "low suitability" and "unsuitable". Development decision assessment report refers to a comprehensive report document generated according to the suitability level to provide a basis for underground space development decisions. The decision assessment report usually includes assessment results, risk analysis, recommended development strategies and related maps.

[0021] Specifically, the suitability level can be determined by directly comparing the underground space development score with a preset grading threshold. For example, a score above 80 can be set as "highly suitable", 60-80 as "medium suitable", and below 60 as "low suitable". The development decision assessment report can be generated using a preset report template. The assessment results, suitability level and general recommendations can be filled into the template to form the final development decision assessment report.

[0022] For example, suppose a coastal city A plans to develop a large-scale underground complex project in its coastal area. The coastal area has a complex geological environment and is frequently affected by typhoons. In this case, traditional assessment methods may be unable to accurately assess its suitability for development.

[0023] First, the target area is identified as the coastal area of ​​the coastal city A, and a multi-dimensional evaluation index system is established, including engineering geological conditions, hydrogeological conditions, adverse geological indicators, and urban development demand indicators as primary indicators. To cope with dynamic factors such as typhoons, a first dynamic evaluation factor is artificially integrated into the hydrogeological conditions, such as parameters like tidal changes and rainfall during historical typhoons. At the same time, a second dynamic evaluation factor is artificially integrated into the adverse geological indicators, such as parameters like land subsidence and soil liquefaction risk caused by historical typhoons.

[0024] Furthermore, multi-source datasets of the target area are acquired in real time, including geological exploration reports from geological management departments, meteorological monitoring data from meteorological management departments, and urban status maps and planning management documents from urban planning management departments. At this point, due to the different data formats and inconsistent timestamps and spatial coordinates, spatiotemporal alignment processing is required. For example, the spatial information in the CAD drawings is converted into a unified geographic coordinate system, and all data is uniformly labeled with timestamps, such as organizing all data into quarterly units. After the above spatiotemporal alignment processing, a standardized dataset is generated.

[0025] Furthermore, based on standardized datasets, the primary evaluation weights of each primary indicator are determined through a pre-defined combined weighting strategy, and the grading thresholds of each primary indicator are determined through a pre-trained grading threshold optimization model. For example, the analytic hierarchy process (AHP) can be used to organize the experience of experts in geology, hydrology, and planning to compare engineering geological conditions, hydrogeological conditions, adverse geological indicators, and urban development demand indicators, and calculate the subjective weights of each indicator. At the same time, standards such as urban underground space development and utilization requirements can be referenced to set suitability level classification thresholds for each indicator. For example, indicators such as groundwater level depth and foundation bearing capacity can be divided into four levels: "excellent," "good," "medium," and "poor," and the corresponding numerical ranges can be given.

[0026] Furthermore, based on the determined primary assessment weights and grading thresholds, the standardized dataset is weighted and fused to generate an underground space development score for the target area. Specifically, for each geographic grid unit within the coastal area, the values ​​of each indicator in its standardized dataset are multiplied by the corresponding primary assessment weight, and then linearly weighted and summed to obtain the comprehensive score for that grid unit. By calculating for all grid units, an underground space development score distribution map of the entire coastal area can be obtained.

[0027] Finally, the suitability level of the target area is determined based on the underground space development score, and a development decision assessment report corresponding to the suitability level is generated. For example, according to the preset scoring threshold, the grid unit with a score of 75.5 is classified as "moderately suitable". By classifying the scores of the entire area, a suitability level distribution map can be obtained. Based on this, a development decision assessment report is generated. This development decision assessment report details the suitability level of each part of the coastal area, the main influencing factors, and the preliminary development recommendations for different levels of areas. For example, it is recommended to give priority to the development of "highly suitable" areas, to conduct further exploration and risk assessment for "moderately suitable" areas, and to develop "lowly suitable" areas with caution or avoid development.

[0028] Compared to existing technologies that rely excessively on static parameters and historical average data for assessment, this embodiment integrates a first dynamic assessment factor and a second dynamic assessment factor into hydrogeological conditions and adverse geological indicators, respectively, which can more comprehensively capture the changes in the complex dynamic geological environment of coastal cities. For example, in the above example, by integrating parameters such as tidal changes and rainfall during typhoons as the first dynamic assessment factor, and parameters such as land subsidence and soil liquefaction risk caused by typhoons as the second dynamic assessment factor, the assessment system can initially consider the impact of extreme meteorological events such as typhoons on the suitability of underground space development, thereby avoiding the problem of the dynamic risk factors being seriously underestimated in traditional methods.

[0029] Furthermore, existing technologies often limit the setting of key evaluation parameters to a single algorithm output or expert subjective judgment. This embodiment, however, determines the first-level evaluation weights through a combined weighting strategy and uses a pre-trained grading threshold optimization model to determine the grading thresholds. In the above example, the weights are determined by combining the analytic hierarchy process with expert scoring, and the grading thresholds are set with reference to local requirements and expert experience. Although there are still subjective elements, the combination strategy and the introduction of model pre-training make the parameter setting more objective and accurate than purely subjective judgment.

[0030] Meanwhile, in this embodiment, real-time acquisition and spatiotemporal alignment of multi-source datasets are performed to generate standardized datasets. In the example above, heterogeneous data such as geological exploration data, meteorological monitoring data, urban status data, and planning management data are initially integrated and standardized through transformation and unified labeling, providing a unified data foundation for weighted fusion calculation, overcoming the shortcomings of existing technologies in the fusion of multi-source heterogeneous data, and improving the accuracy and reliability of evaluation results.

[0031] In summary, this embodiment introduces dynamic evaluation factors, adopts a combined weighting strategy and a hierarchical threshold optimization model, and performs spatiotemporal alignment processing on multi-source heterogeneous data to form a more comprehensive and dynamic-adaptable method for assessing the suitability of underground space resource development. This method can solve problems such as insufficient dynamic risk assessment, unreasonable parameter settings, and lack of data fusion capabilities in existing technologies. It has the effect of improving the accuracy and real-time performance of assessments and reducing resource development risks under extreme events such as typhoon seasons.

[0032] In one embodiment, such as Figure 2 As shown, step S10 includes: S11: Engineering geological conditions, hydrogeological conditions, adverse geological indicators, and urban development demand indicators are used as primary indicators in the multi-dimensional evaluation indicators. In this embodiment, engineering geological conditions, hydrogeological conditions, adverse geological indicators, and urban development demand indicators are used as primary indicators in the multi-dimensional evaluation system. These four primary indicators serve as a macro-classification for assessing the suitability of underground space, covering the main directions affecting underground space development. Among them, engineering geological conditions can focus on strata lithology, structure, stability, bearing capacity, etc.; hydrogeological conditions can focus on groundwater occurrence, movement, hydrochemical properties, recharge and discharge conditions, etc.; adverse geological indicators can focus on disaster risks such as earthquakes, landslides, collapses, and ground fissures; and urban development demand indicators can focus on planning, economic, social, and environmental needs.

[0033] S12: Obtain the technical standards for the primary indicators and the development requirements for the target area, and decompose each primary indicator into its corresponding set of secondary indicators based on the technical standards for the primary indicators and the development requirements for the target area. In this embodiment, the technical standards for indicators refer to the standards formulated by the industry or locality for various geological, hydrological, and engineering conditions; the development requirements of the target area refer to the specific needs of a particular project or area for the function, depth, scale, safety level, and environmental impact of underground space development; by combining the technical standards corresponding to the primary indicators and the development requirements of the target area, the macro-level primary indicators can be refined into more specific secondary indicators. For example, hydrogeological conditions can be decomposed into secondary indicators such as groundwater level depth, permeability coefficient, groundwater corrosivity, and tidal influence range; engineering geological conditions can be decomposed into secondary indicators such as stratigraphic lithology, soil strength, foundation bearing capacity, and site stability.

[0034] S13: Obtain the first historical dataset of the target area, and based on the first correlation model and the first historical dataset, screen out typhoon impact parameters from the secondary indicators of hydrogeological conditions, and integrate them into the first dynamic evaluation factor. In this embodiment, the first historical dataset refers to a collection of data related to hydrogeological conditions and typhoon activity in the target area over a past period, such as historical rainfall, groundwater level monitoring data, tidal data, river runoff, typhoon path, intensity, landfall point, and duration. The first correlation model refers to a mathematical model that can reveal the intrinsic relationship between typhoon activity and hydrogeological conditions, such as a model based on machine learning or statistical regression. By analyzing the first historical dataset through the first correlation model, secondary indicators that show significant changes in hydrogeological conditions under the influence of typhoons can be identified, which are the typhoon impact parameters. These parameters are then integrated into the first dynamic evaluation factor to reflect the dynamic changes in hydrogeological conditions.

[0035] S14: Obtain the second historical dataset of the target area, and based on the second correlation model and the second historical dataset, screen out typhoon disturbance parameters from the secondary index set of adverse geological indicators, and integrate them into the second dynamic evaluation factor.

[0036] In this embodiment, the second historical dataset refers to a collection of data related to adverse geological indicators and typhoon activity in the target area over a past period, such as historical geological disaster records, soil moisture content, surface deformation monitoring data, changes in groundwater chemical composition, typhoon path, intensity, landfall point, and duration. The second correlation model refers to a mathematical model that can reveal the intrinsic relationship between typhoon activity and adverse geological indicators, such as a model based on machine learning or statistical analysis. By analyzing the second historical dataset through the second correlation model, secondary indicators that show significant risk or incidence changes in adverse geological indicators under typhoon disturbances can be identified, which are the typhoon disturbance parameters. These parameters are then integrated into a second dynamic assessment factor to reflect the dynamic risk of adverse geological indicators.

[0037] For example, as a specific implementation method, when determining the target area and establishing multi-dimensional evaluation indicators, engineering geological conditions, hydrogeological conditions, adverse geological indicators, and urban development demand indicators can be identified as primary indicators of the evaluation system. For instance, for a coastal city, its underground space development requirements may include "seismic fortification intensity not less than 8 degrees" and "groundwater control level not higher than 5 meters below ground level." Combining building foundation design requirements, urban flood control standards, and other technical standards, engineering geological conditions can be decomposed into secondary indicators such as foundation bearing capacity, soil liquefaction potential, and site stability; hydrogeological conditions can be decomposed into secondary indicators such as groundwater depth, permeability coefficient, and groundwater corrosivity; adverse geological indicators can be decomposed into secondary indicators such as active fault distribution, ground subsidence rate, and landslide susceptibility; and urban development demand indicators can be decomposed into secondary indicators such as population density, transportation demand, and industrial layout. When integrating dynamic evaluation factors, historical typhoon information for the city over the past 20 years, as well as hydrological data such as groundwater level, tide level, and soil moisture content during the same period, can be obtained. Geological monitoring data constitutes the first historical dataset. Simultaneously, this data is used to train a first association model based on a long short-term memory network to predict the changing trends of secondary indicators such as groundwater level depth and permeability coefficient under different typhoon conditions. By analyzing the model's prediction results, secondary indicators with predicted change rates exceeding 10% under typhoon influence, such as groundwater level depth, are selected and integrated as typhoon impact parameters, forming the first dynamic evaluation factor. Similarly, historical typhoon information for the city over the past 20 years, along with concurrent surface deformation monitoring data and geological disaster records, constitutes the second historical dataset. This data is used to train a second association model based on a support vector machine to predict the probability of occurrence or risk change rate of adverse geological indicators such as ground subsidence rate and collapse risk under different typhoon conditions. By analyzing the model's prediction results, secondary indicators with predicted occurrence rates or risk change rates exceeding preset risk thresholds under typhoon influence, such as ground subsidence rate, are selected and integrated as typhoon disturbance parameters, forming the second dynamic evaluation factor.

[0038] Through the above technical solution, this application can incorporate the unique dynamic environmental factors of coastal cities into a multi-dimensional evaluation index system, especially the impact of typhoons on the suitability of underground space development. Specifically, by decomposing the primary indicators, using correlation models and corresponding historical data to screen typhoon impact parameters and typhoon disturbance parameters, and integrating them into dynamic evaluation factors, the limitations of traditional evaluation methods in not considering dynamic environmental factors can be overcome. This allows the evaluation results to more accurately reflect the true suitability of underground space in coastal cities under complex and changeable environments, thus improving the scientificity and reliability of the evaluation.

[0039] In one embodiment, such as Figure 3 As shown, step S13 includes: S131: Obtain historical typhoon information of the target area and its corresponding first time series data to form a first historical dataset. The first time series data includes a secondary index set of engineering geological conditions, a secondary index set of hydrogeological conditions, and typhoon characteristic parameters. In this embodiment, historical typhoon information refers to detailed records of typhoon events that have occurred in the target area or its influence range in the past. It typically includes the typhoon's path, intensity, speed, duration, and landfall location. Its purpose is to provide background and characteristic data of the typhoon events for subsequent analysis. This historical typhoon information can be obtained from historical meteorological databases of official agencies such as meteorological management departments and marine management departments, or through historical retrospective analysis of satellite remote sensing data and radar data. This is clear to those skilled in the art. The first time-series data refers to a continuous data sequence related to typhoon events collected at different points in time, reflecting the situation before, during, and after the typhoon. The function of this data set is to capture the evolution of engineering geological conditions, hydrogeological conditions, and typhoon characteristics over time. First-time series data can be continuously collected through a long-term deployed sensor network, or obtained through periodic manual measurements or remote sensing image analysis to acquire geological and hydrological parameters at different time points. Second-level indicators for engineering geological conditions refer to a detailed set of parameters that refine engineering geological conditions, such as foundation bearing capacity, soil strength, slope stability coefficient, stratum lithology, and fault structure distribution. These serve as input features for model training, helping to understand the impact of typhoons on geological conditions. The second-level indicators for engineering geological conditions can be obtained from geological survey reports. Data can be obtained through drilling, indoor and outdoor testing, or geophysical exploration to obtain information on underground structures. The secondary index set of hydrogeological conditions refers to a detailed set of parameters that refine hydrogeological conditions, such as groundwater level depth, aquifer permeability coefficient, pore water pressure, groundwater flow velocity, surface runoff coefficient, and rainfall infiltration rate. Its function is to serve as the prediction target for model training, revealing the degree of typhoon impact on hydrogeological conditions. This secondary index set can be obtained through groundwater monitoring wells, hydrological station data, or through hydrogeological surveys and pumping tests. Typhoon characteristic parameters refer to key indicators used to describe the typhoon's own attributes, such as typhoon central pressure, maximum wind speed, and typhoon... Radius, movement path, rainfall, etc., serve as input features for model training, quantifying the intensity and impact range of typhoon events. These typhoon characteristic parameters can be extracted from meteorological stations and satellite observation data, or simulated through numerical weather prediction models. The first historical dataset refers to the data set formed by integrating and aligning the aforementioned historical typhoon information and first time-series data along the time dimension. Its purpose is to provide sufficient and time-series-related sample data for subsequent machine learning model training. This can be achieved by cleaning, unifying the format, and aligning the timestamps of data from different sources before storing them in a relational database or time-series database, or by using data warehouse technology to integrate multi-source heterogeneous data.

[0040] S132: Using the secondary index set of typhoon characteristic parameters and engineering geological conditions as input, and the secondary index set of hydrogeological conditions as the prediction target, the first historical dataset is trained through the gradient boosting tree model to generate the first association model. In this embodiment, the gradient boosting tree model refers to an ensemble learning algorithm that iteratively trains several weak predictors and accumulates their results to minimize the loss function. Its characteristics include the ability to handle nonlinear relationships, low sensitivity to outliers, and high prediction accuracy. Specifically, the gradient boosting tree model constructs a prediction model capable of capturing the complex nonlinear relationships between typhoon characteristic parameters, engineering geological conditions, and hydrogeological conditions. Specifically, the gradient boosting tree model can be implemented using open-source libraries such as XGBoost, LightGBM, or CatBoost, and the model's hyperparameters, such as the learning rate, the number of trees, and the tree depth, can be adjusted according to the characteristics of the actual data. The first correlation model refers to the mathematical model obtained after training with the gradient boosting tree model, which reveals the quantitative relationship between the secondary index sets of typhoon characteristic parameters, engineering geological conditions, and hydrogeological conditions. Its function is to predict changes in hydrogeological conditions given typhoon conditions and engineering geological conditions.

[0041] S133: Input the preset typhoon conditions and their corresponding engineering geological conditions into the first correlation model, and calculate the predicted change rate of the secondary index set of engineering geological conditions and the secondary index set of hydrogeological conditions under the preset typhoon conditions respectively. In this embodiment, the preset typhoon condition characteristics refer to the characteristic parameters that simulate future typhoon events, such as specific intensity, path, and rainfall. These can be set based on historical extreme events, climate model predictions, or engineering design standards. The characteristic values ​​of engineering geological conditions refer to the current or expected values ​​of the secondary indicators of engineering geological conditions in the target area under the preset typhoon conditions. The predicted rate of change refers to the expected change of the secondary indicators of engineering geological conditions and hydrogeological conditions relative to their normal or baseline states under specific typhoon conditions. These can be percentage changes, absolute value changes, or grade changes.

[0042] S134: Secondary indicators whose predicted rate of change exceeds a preset sensitivity threshold are identified as typhoon impact parameters, and typhoon impact parameters are integrated into the first dynamic evaluation factor.

[0043] In this embodiment, the preset sensitivity threshold is a pre-set value used to determine the sensitivity of secondary indicators to the impact of typhoons. For example, it can be set to a change rate of 5% or 10%. When the predicted change rate of a certain secondary indicator exceeds this threshold, it is considered that the secondary indicator is significantly affected by the typhoon. The typhoon impact parameter refers to the secondary indicators identified as having significant sensitivity to typhoon events, which directly reflects the key impact of typhoons on hydrogeological and engineering geological conditions. The first dynamic assessment factor is a comprehensive factor integrated from the typhoon impact parameters, used to dynamically reflect the impact of typhoons on hydrogeological conditions in the suitability assessment of underground space development.

[0044] For example, as a specific implementation method, data on the paths, intensities, and rainfall of all typhoons that have passed through the target area in the past 30 years can be obtained from the meteorological center. Simultaneously, continuous monitoring data synchronized with the typhoon events can be obtained from groundwater level monitoring stations, pore water pressure sensors, surface runoff monitoring points, and soil moisture sensors within the target area, forming a first historical dataset. For instance, during the passage of a certain typhoon, its maximum wind speed was recorded as 50 m / s, its central pressure as 930 hPa, and data such as a 2-meter rise in groundwater level and a 50 kPa increase in pore water pressure were collected. A gradient boosting tree model is constructed using the XGBoost library in Python, using historical typhoon maximum wind speed, total rainfall, and typhoon center distance as typhoon characteristic parameters, and foundation bearing capacity and soil stability as secondary indicators of engineering geological conditions. Groundwater level, pore water pressure, and permeability coefficient are used as secondary indicators of hydrogeological conditions. As the prediction target, model parameters are optimized through cross-validation and grid search. For example, the number of trees can be set to 1000, the maximum depth to 6, and the learning rate to 0.01. Assuming a relatively severe typhoon condition, characterized by a maximum wind speed of 60 m / s and rainfall of 500 mm / 24h, the above features and the engineering geological conditions of the current target area are input into the trained first association model to output prediction results, such as predicting that the groundwater level will rise by 3 meters and the pore water pressure will increase by 70 kPa. Then, the percentage change of the predicted value relative to the normal state is calculated. For example, if the groundwater level rises from the normal 10 meters to 13 meters, the change rate is 30%. The sensitivity threshold is set to 15%. If the predicted change rate of the groundwater level is 30%, then the groundwater level is determined as a typhoon impact parameter. If the predicted change rate of the pore water pressure is 20%, then the pore water pressure is also determined as a typhoon impact parameter. Finally, the determined parameters are integrated into the first dynamic evaluation factor.

[0045] By incorporating historical typhoon information and time-series data, and constructing a first correlation model using a gradient boosting tree model, this application can accurately capture the complex nonlinear dynamic relationship between typhoon events and hydrogeological and engineering geological conditions. This allows for the quantitative prediction of the change rate of various secondary indicators when facing preset typhoon conditions, and the selection of key parameters truly affected by typhoons based on sensitivity thresholds. Consequently, the integrated first dynamic evaluation factor can more accurately and timely reflect the dynamic impact of typhoons on the suitability of underground space development in coastal cities, improving the scientificity and reliability of the evaluation results, and avoiding potential risks caused by static evaluation.

[0046] In one embodiment, step S14 includes: S141: Obtain historical typhoon information of the target area and its corresponding second time series data to form a second historical dataset. The second time series data includes a secondary index set of adverse geological indicators, a secondary index set of hydrogeological conditions, and typhoon characteristic parameters. In this embodiment, the second historical dataset refers to historical observation data used to train and validate the model. It contains various geological, hydrological, and meteorological information related to typhoon activity in the target area at different time points, aiming to provide sufficient samples for training the correlation model to reveal the potential connection between typhoon activity and adverse geological phenomena. The second historical dataset may include long-term monitoring data, historical disaster records, remote sensing image data, etc. The second time series data is a specific component of the second historical dataset. It records the secondary indicator set of adverse geological indicators, the secondary indicator set of hydrogeological conditions, and typhoon characteristic parameters in the form of time series. It can reflect the changing trend of indicators over time and their dynamic response before, during, and after typhoon events. For example, the second time series data may include data on groundwater level, pore water pressure, soil moisture content, surface subsidence, and crack development before and after the passage of a typhoon.

[0047] S142: Using the secondary index set of typhoon characteristic parameters and hydrogeological conditions as input, and the secondary index set of adverse geological indicators as the prediction target, the second historical dataset is trained by a logistic regression model to generate a second correlation model. In this embodiment, the logistic regression model is a generalized linear model, often used to solve binary or multi-class classification problems. By mapping the results of linear regression to the interval (0, 1), it outputs the probability of an event occurring. In this embodiment, its function is to establish a complex nonlinear relationship between typhoon characteristic parameters, hydrogeological conditions, and adverse geological indicators to predict the probability of adverse geological events or the rate of risk change under specific typhoon conditions. Furthermore, in addition to the logistic regression model, classification models such as support vector machines or neural networks can also be used to achieve similar functions.

[0048] S143: Input the preset typhoon conditions and their corresponding hydrogeological conditions into the second correlation model to calculate the predicted occurrence rate or risk change rate of the secondary index set of adverse geological indicators under the preset typhoon conditions. In this embodiment, the second association model is obtained by training a logistic regression model on a second historical dataset. It learns and quantifies the association strength and patterns between typhoon characteristics, hydrogeological conditions, and adverse geological indicators, thereby predicting the response of adverse geological indicators under given input conditions. The second association model can be a parameter-optimized logistic regression classifier or other trained machine learning models, such as decision tree models or gradient boosting models. The preset typhoon condition characteristics refer to representative typhoon event scenarios set during the assessment, such as typhoons of specific intensity, path, rainfall, and duration, used to simulate possible future typhoon events to assess their potential impact on adverse geological indicators. These typhoon condition characteristics can be based on the strongest historical typhoon, the average typhoon, or typhoons of a specific risk level. The model can be set as a future typhoon scenario based on climate change predictions. The characteristic values ​​of hydrogeological conditions refer to specific values ​​related to hydrogeological conditions under preset typhoon conditions, such as groundwater level, pore water pressure, and soil saturation. These are used as inputs to the second correlation model and, together with the typhoon condition characteristics, determine the predicted response of adverse geological indicators. The characteristic values ​​of hydrogeological conditions can be obtained through hydrological model simulation or typical values ​​based on historical data statistical analysis. The predicted occurrence rate or risk change rate refers to the result calculated by the second correlation model based on the input typhoon condition characteristics and the characteristic values ​​of hydrogeological conditions. It represents the probability of adverse geological events occurring or the degree of change of their risk level relative to the normal under the influence of a specific typhoon. For example, it can output the percentage probability of geological disasters or the risk index increment of a certain adverse geological phenomenon.

[0049] S144: Secondary indicators that predict the occurrence rate or risk change rate exceeds the preset risk threshold are identified as typhoon disturbance parameters, and typhoon disturbance parameters are integrated into a second dynamic assessment factor.

[0050] In this embodiment, the preset risk threshold is a predetermined critical value used to determine whether the predicted occurrence rate or risk change rate has reached a level requiring attention. When the predicted value exceeds the risk threshold, the corresponding secondary indicator is considered to be significantly disturbed by the typhoon. The risk threshold can be set according to engineering experience, safety specifications, or risk management strategies. For example, it can be set to a 50% probability of occurrence or a 20% risk change rate. The typhoon disturbance parameter refers to the secondary indicator of adverse geological indicators whose predicted occurrence rate or risk change rate exceeds the preset risk threshold under preset typhoon conditions. For example, if the liquefaction potential index's predicted occurrence rate exceeds the risk threshold under typhoon conditions, then the liquefaction potential index is the typhoon disturbance parameter. The second dynamic evaluation factor is a comprehensive factor integrated from the screened typhoon disturbance parameters. It represents the part of the adverse geological indicators that is dynamically affected by the typhoon and can more accurately reflect the dynamic risk of underground space in coastal cities under the influence of typhoons. The second dynamic evaluation factor can be a vector containing all typhoon disturbance parameters or a weighted aggregation of typhoon disturbance parameters.

[0051] For example, as a specific implementation method, suppose a coastal city needs to assess the liquefaction and landslide risks of its underground space under the influence of typhoons. First, collect information on the historical typhoon paths, intensities, and rainfall over the past 30 years, along with hydrogeological data such as groundwater level, pore water pressure, and soil moisture content, and adverse geological data such as surface subsidence, cracks, liquefaction event records, and landslide event records, to form a second historical dataset. The second time-series data can specifically include: a set of secondary indicators for adverse geological indicators, such as the liquefaction potential index and landslide sensitivity index; a set of secondary indicators for hydrogeological conditions, such as groundwater depth and pore water pressure; and typhoon characteristic parameters, such as typhoon central pressure, maximum wind speed, and 24-hour cumulative rainfall. Further, a logistic regression model can be constructed using the Scikit-learn library in Python. The typhoon characteristic parameters and the set of secondary indicators for hydrogeological conditions are used as input features of the logistic regression model, and the set of secondary indicators for adverse geological indicators is used as the prediction target. The second historical dataset is then trained to generate a second correlation model. For example, the model can be trained... The model predicts the probability that the liquefaction potential index will exceed a certain critical value under given typhoon and hydrological conditions. It then sets preset typhoon characteristics, such as simulating a strong typhoon making landfall near the coast with a maximum wind speed of 45 m / s and a 24-hour cumulative rainfall of 250 mm, and inputs corresponding hydrogeological characteristics, such as an initial groundwater level of 2 meters below the surface. The input data is then fed into a pre-trained second correlation model to calculate the predicted occurrence rate or risk change rate of the liquefaction potential index and landslide sensitivity index under the current typhoon conditions. For example, the second correlation model might... The predicted occurrence rate of the liquefaction potential index is 75%, and the risk change rate of the landslide sensitivity index is 40%. Finally, a preset risk threshold is set. For example, an occurrence rate exceeding 60% or a risk change rate exceeding 30% is considered a significant disturbance. Based on this risk threshold, if the predicted occurrence rate of the liquefaction potential index is 75%, it is determined as a typhoon disturbance parameter; if the risk change rate of the landslide sensitivity index is 40%, it is also determined as a typhoon disturbance parameter. Finally, the determined typhoon disturbance parameters are integrated into a second dynamic assessment factor for subsequent suitability assessment of underground space development.

[0052] Through the above technical solution, this application can accurately capture the complex dynamic correlation between typhoon activity and adverse geological indicators by utilizing historical data and machine learning models. This effectively solves the limitations of traditional assessment methods in dealing with the impact of extreme weather events. Specifically, by constructing a second historical dataset and training a logistic regression model to generate a second correlation model, it is possible to quantitatively calculate the predicted occurrence rate or risk change rate of adverse geological indicators under preset typhoon conditions. Through this dynamic assessment method based on prediction models, parameters that are truly significantly disturbed by typhoons can be screened from the secondary indicators of adverse geological indicators and integrated into a second dynamic assessment factor. This not only improves the scientificity and accuracy of adverse geological indicator assessment but also enables development suitability assessment to fully consider the dynamic impact of extreme weather events such as typhoons, providing a more forward-looking and reliable assessment basis, thereby effectively reducing the risks that underground space development may face.

[0053] In one embodiment, step S20 includes: S21: Use deep learning OCR to parse unstructured data in multi-source datasets, extract effective information, and transform it into semi-structured data; In this embodiment, deep learning OCR refers to the technology of recognizing text in images using deep neural networks. Its function is to automatically identify and extract text information from unstructured data and convert it into a machine-readable and processable semi-structured format, such as JSON, XML, or text with specific tags, for data processing and analysis. Deep learning OCR can use an end-to-end OCR model based on a combination of convolutional neural networks and recurrent neural networks, such as CRNN or Transformer-based OCR models, to detect and recognize text regions in images or PDFs. Alternatively, it can utilize pre-trained OCR service APIs, such as AI open platforms, to upload unstructured documents to the service for parsing and obtain structured text output.

[0054] S22: Based on a preset unified spatial reference coordinate system, coordinate transformation and mapping matching are performed on spatial data in multi-source datasets to achieve spatial dimension alignment; In this embodiment, spatial data typically uses different geographic coordinate systems or projected coordinate systems. A unified spatial reference coordinate system refers to a standardized coordinate system pre-determined before data processing, such as WGS84, CGCS2000, or a specific region's UTM projection. All spatial data will be transformed to this coordinate system. The role of coordinate transformation and mapping matching is to eliminate spatial positional deviations between different data sources, ensuring that all geographic information is overlaid and analyzed within the same spatial framework. This can be achieved using coordinate transformation tools provided by geographic information system software, by defining source and target coordinate systems, and performing projection transformations or reference surface transformations. Alternatively, batch coordinate transformation operations can be implemented using programming languages ​​combined with geospatial processing libraries, and the transformed data can be verified and matched for spatial topological relationships.

[0055] S23: Based on the preset time granularity, normalize the time series data in the multi-source dataset, label the non-time series data with a unified timestamp, and use interpolation to fill in the gaps in the data sequence to achieve time dimension alignment; In this embodiment, since multi-source data may be collected at different time frequencies or irregular time intervals, some data may not have explicit timestamps. Time granularity regularization refers to unifying all time-series data to a preset consistent time interval, such as hourly or daily. Labeling non-time-series data with a unified timestamp is to incorporate it into the time series analysis framework. Interpolation to fill gaps is to solve the problem of missing data, ensure the integrity of the time series, and thus achieve synchronization and comparability of all data in the time dimension. For time-series data, resampling techniques can be used, such as downsampling high-frequency data to low-frequency data, or upsampling low-frequency data to high-frequency data through linear interpolation, spline interpolation, etc. For non-time-series data, a unified timestamp can be assigned according to its collection or update time, or according to the time of occurrence of its associated events. For data gaps, various interpolation methods can be used, such as linear interpolation, polynomial interpolation, spline interpolation, nearest neighbor interpolation, or more complex time series prediction models based on machine learning for filling gaps.

[0056] S24: The 3σ criterion is used to verify the multi-source dataset after spatial and temporal alignment, identify outlier data points and correct them to obtain the outlier processing results. In this embodiment, the 3σ criterion is a statistical method used to identify outliers in the data. Specifically, it assumes that the data follows a normal distribution. If a data point deviates from the mean by more than three times the standard deviation, it is considered an outlier. Its purpose is to further improve data quality after data alignment, eliminate data points caused by obvious errors or measurement errors, and thus improve the accuracy of the evaluation. Correcting outliers can prevent them from having an undue impact on the overall analysis results. For identified outliers, a substitution method can be used for correction, such as replacing them with the mean, median, mode, or interpolation results of adjacent points in the data sequence containing the outlier. Alternatively, more complex outlier handling strategies can be used, such as machine learning-based anomaly detection algorithms, such as Isolation Forest and LOF. After identifying outliers, they can be reviewed and corrected according to business logic or expert experience, or they can be directly marked as invalid data.

[0057] S25: Based on the outlier processing results, exclude invalid data in the multi-source dataset whose deviation exceeds the preset deviation threshold, and perform unified processing on the remaining data after exclusion to generate a standardized dataset.

[0058] In this embodiment, the outlier handling results have already provided a preliminary judgment on data quality. Based on this, a deviation threshold is further set to exclude invalid data that, although it may not be statistically significant outliers, clearly does not meet business requirements or engineering specifications in terms of value or quality. Unified processing refers to standardizing all cleaned, aligned, and filtered data in terms of format, units, and dimensions to make it consistent and comparable, ultimately forming a standardized dataset that can be directly used to evaluate the model. The deviation threshold can be set according to industry standards, expert experience, or historical data distribution characteristics. For example, for a certain indicator, if its value exceeds the reasonable range in terms of physics or engineering, it can be considered invalid data even if it does not meet the outlier standard of the 3σ criterion. Unified processing can include data normalization, standardization, or dimension unification. After excluding invalid data, the remaining data can be subjected to data type conversion, encoding, feature engineering, and other operations to adapt to the input requirements of the evaluation model. Generating a standardized dataset usually means that the data has been cleaned, transformed, integrated, and stored in a unified format, such as relational database tables or data frames.

[0059] Specifically, to address the prevalent unstructured information in multi-source data, deep learning OCR technology is used for analysis. This accurately extracts key information from unstructured data such as images and scanned documents, transforming it into semi-structured data. This breaks down data format barriers, allowing information that would otherwise be difficult for machines to process directly to be included in the evaluation. Furthermore, to resolve the issue of inconsistent spatial locations across different data sources, a pre-defined unified spatial reference coordinate system is used to perform coordinate transformation and mapping matching on all spatial data, ensuring that all geographic information is overlaid within the same spatial framework. Simultaneously, considering the differences in data collection frequency and timestamps, this solution regularizes time-series data using a pre-defined time granularity and labels non-time-series data with unified timestamps. Interpolation is then used to fill gaps in the data sequence, achieving alignment in the time dimension and ensuring the comparability and sequence integrity of data at different time points. After initial spatiotemporal alignment, to further improve data quality, the 3σ criterion is used to verify the data, identifying and correcting outlier data points. This effectively eliminates interference from measurement errors or data entry errors. Based on this, combined with outlier processing results, invalid data with deviations exceeding a pre-defined threshold is further eliminated to ensure data reliability.

[0060] Through the above technical solutions, this application can effectively solve the inconsistency problems of multi-source heterogeneous data in terms of format, space, time and quality. Specifically, through deep learning OCR technology, unstructured data can be parsed, data sources can be broadened and data utilization can be improved. At the same time, a unified spatial reference coordinate system and temporal granularity regularization can ensure the alignment of different data sources in the spatiotemporal dimension and eliminate obstacles to data fusion. The 3σ criterion combined with the deviation threshold outlier and invalid data processing mechanism can improve the purity and reliability of the data.

[0061] In one embodiment, step S30 includes: S31: The objective value weights of each primary indicator are calculated using the entropy weight method, and the subjective experience weights of each primary indicator are obtained through a pre-set hierarchical analysis strategy. In this embodiment, the entropy weight method is an objective weighting method. Its basic principle is to determine the weight based on the degree of variation of the indicator value. The smaller the information entropy of the indicator, the greater the degree of variation of the indicator value, the more information it provides, and the greater its weight. Conversely, the larger the information entropy, the smaller the degree of variation of the indicator value, the less information it provides, and the smaller its weight. The entropy weight method can extract the relative importance of each indicator from the data itself, avoiding the bias of subjective judgment. Another implementation method is to use the CRITIC method, which determines the objective weight by considering the comparative strength of the indicators and the conflict between the indicators, which can more comprehensively reflect the objective importance of the indicators. The preset hierarchical analysis strategy is a multi-criteria decision-making method that combines qualitative and quantitative analysis. It decomposes complex problems into several levels and several factors, and compares each factor pairwise to determine the relative importance of each factor, and finally obtains the subjective experience weight. This hierarchical analysis strategy can effectively integrate the knowledge and experience of domain experts and reflect the preferences of decision-makers. Another implementation method is to use the Delphi method, which gradually converges expert opinions through multiple rounds of anonymous expert consultation, thereby obtaining the subjective experience weight of each first-level indicator to reduce the influence of individual expert subjectivity.

[0062] S32: Obtain standardized indicator data and development suitability levels of historical engineering cases, and use the obtained standardized indicator data and development suitability levels as training features and training labels respectively to construct the first training sample set; In this embodiment, the standardized indicator data of historical engineering cases refers to the values ​​of various evaluation indicators corresponding to previously completed underground space development projects. These indicators have been standardized to eliminate the influence of dimensions and orders of magnitude, making them comparable. The development suitability level refers to the classification result of the suitability of underground space development determined by historical cases after actual verification or expert evaluation, such as "very suitable", "suitable", "generally suitable", "unsuitable", etc. The first training sample set refers to the data set composed of standardized indicator data of historical engineering cases as training features and development suitability levels corresponding to historical engineering cases as training labels. The first training sample set is used for supervised learning, aiming to train the model to learn the mapping relationship from indicator data to suitability levels.

[0063] S33: Train the random forest model using the first training sample set to generate a weight correction model, and input the linear combination result of objective value weights and subjective experience weights into the weight correction model to obtain dynamic correction coefficients; In this embodiment, the random forest model is an ensemble learning method that makes a final prediction by constructing multiple decision trees and voting or averaging the prediction results. The random forest model has advantages such as strong resistance to overfitting, good performance in handling high-dimensional data, and the ability to handle nonlinear relationships, making it suitable for complex classification or regression tasks. Another machine learning model that can be used to train the weight correction model is the support vector machine, which separates samples of different classes by finding an optimal hyperplane in a high-dimensional space. It has good generalization ability for small sample sizes, nonlinearity, and high-dimensional pattern recognition problems. The weight correction model refers to the random forest model trained on the first training sample set. Its function is to adjust and optimize the initially determined indicator weights based on historical experience and actual results. Among them, the weight correction model can learn the impact of different indicator combinations on the final suitability assessment results, thereby providing a data-driven correction mechanism. The linear combination result of objective value weight and subjective experience weight refers to the initial weight obtained by weighting the objective value weight obtained through the entropy weight method and the subjective experience weight obtained through the analytic hierarchy process according to a certain ratio. Its purpose is to take into account both the objectivity of the data and the subjective judgment of experts to form a comprehensive initial weight. For example, the two can be equally weighted and averaged, or different weight coefficients can be assigned according to actual needs for combination.

[0064] S34: Based on the dynamic correction coefficient, the linear combination result of objective value weight and subjective experience weight is dynamically adjusted to output the first-level evaluation weight of each first-level indicator.

[0065] In this embodiment, the dynamic correction coefficients are adjustment factors output by the weight correction model based on the linear combination of the input objective value weights and subjective experience weights. These coefficients are used to refine the initial combined weights, making them more consistent with actual assessment results and historical experience. Dynamic adjustment refers to real-time, adaptive correction of the linear combination of objective value weights and subjective experience weights based on the obtained dynamic correction coefficients. This adjustment can be multiplicative, that is, multiplying the combined weights by the corresponding correction coefficients; or additive, that is, adding or subtracting correction values ​​from the combined weights. The primary assessment weights are the weights of each primary indicator in the multidimensional assessment indicators used for assessment, ultimately determined after dynamic adjustment. These weights integrate objective data, subjective experience, and historical case verification, possessing higher accuracy and adaptability, and can more reliably reflect the relative importance of each primary indicator in the suitability assessment of underground space development.

[0066] For example, as a specific implementation method, when determining the primary evaluation weights of the four primary indicators—engineering geological conditions, hydrogeological conditions, adverse geological indicators, and urban development demand indicators—the entropy weight method can first be used to analyze the standardized dataset of the target area. For instance, if the values ​​of a certain primary indicator fluctuate significantly across different regions, indicating rich information content, the entropy weight method will assign it a higher objective value weight. Simultaneously, based on expert experience, a judgment matrix can be constructed using the analytic hierarchy process (AHP) to compare these four primary indicators pairwise. Furthermore, historical engineering cases of underground space development projects in coastal cities over the past decade are collected, including standardized indicator data for each project and its final development suitability level, and compiled into the first training sample set. Finally, a random forest can be constructed using the scikit-learn library in Python. The classifier serves as the weight correction model and is trained using the first training sample set. During training, hyperparameters such as the number of trees and maximum depth of the random forest can be adjusted to optimize the model's predictive performance. When it is necessary to determine the primary evaluation weight of the current target area, the objective value weight calculated by the entropy weight method and the subjective experience weight obtained by the analytic hierarchy process are linearly combined. The combined weight is then input into the trained random forest weight correction model, which outputs a set of dynamic correction coefficients based on its learned historical experience. For example, if the weight correction model finds that the evaluation results are more accurate when the hydrogeological condition weight is higher in historical cases, it may output a correction coefficient that slightly increases the hydrogeological condition weight. Finally, the dynamic correction coefficients are applied to the preliminary combined weights, for example, by multiplication or addition, to obtain the final primary evaluation weight.

[0067] Through the above technical solution, this application can effectively combine objective data analysis with expert subjective experience when determining the first-level evaluation weights of each first-level indicator in the multi-dimensional evaluation index, thereby avoiding the limitations of a single weighting method. By introducing standardized indicator data from historical engineering cases and developing suitability levels, and using a random forest model to train a weight correction model, dynamic correction of the initial combined weights can be achieved. This ensures that the weight evaluation not only has a data foundation and professional experience support, but also can learn and optimize itself based on historical cases, thereby improving the accuracy, reliability, and dynamic adaptability of the evaluation weights.

[0068] In one embodiment, step S33 includes: S331: Using standardized index data from the first training sample set as input features and the development suitability level corresponding to the standardized index data as the prediction target, supervised learning is performed on the first training sample set using the random forest algorithm. In this embodiment, through supervised learning, the random forest model can learn the inherent mapping relationship from the known inputs and outputs. In addition to the random forest algorithm, other ensemble learning algorithms can also be used, such as gradient boosting decision trees or XGBoost. These methods also combine multiple weak learners to build strong learners to improve prediction performance.

[0069] S332: Optimize the hyperparameters of the random forest model, such as the number of trees, maximum depth, and minimum number of samples per leaf node, using grid search. Train and generate a weighted correction model with the evaluation accuracy of historical cases as the objective function. In this embodiment, grid search is a commonly used hyperparameter tuning method that finds the optimal model configuration by trying predefined parameter combinations. Using the evaluation accuracy of historical cases as the objective function can ensure the effectiveness of the weight correction model in practical applications. In addition to grid search, random search can also be used, which randomly samples a certain number of combinations in the hyperparameter space for evaluation. It is usually more efficient than grid search, especially when the hyperparameter space has a high dimension.

[0070] S333: The objective value weight and the subjective experience weight are weighted and summed to form a basic weight combination. The basic weight combination is then input into the weight correction model to generate the preliminary correction coefficients for the weights of each primary indicator. In this embodiment, the basic weights are initially adjusted using a trained weight correction model to generate preliminary correction coefficients that better reflect the actual situation. The combination of basic weights can be obtained by linear weighted summation; or, a more complex nonlinear combination method can be used, such as fusing objective and subjective weights through a neural network layer to capture more complex interactions.

[0071] S334: Determine the target historical case and apply the preliminary correction coefficient to the target historical case for verification. If the verification result is successful, output the preliminary correction coefficient as the dynamic correction coefficient.

[0072] In this embodiment, the effectiveness of the correction coefficients can be evaluated by validating them on independent historical cases, avoiding the problem that the model performs well on training data but has poor generalization ability on new data. Specifically, K-fold cross-validation can be used, dividing the historical case dataset into training and validation sets. The model is trained on the training set to generate preliminary correction coefficients, which are then evaluated on the validation set. Alternatively, a portion of historical cases not used in model training can be reserved as a test set, and the preliminary correction coefficients can be applied to this test set. The validation is then judged based on preset pass criteria, such as assessing the accuracy reaching a certain threshold. Furthermore, expert experience can be incorporated to qualitatively evaluate and judge the application effect of the preliminary correction coefficients on specific historical cases.

[0073] For example, as a specific implementation method, suppose we have collected detailed data on 1000 underground space development projects in a coastal city over the past 20 years, including standardized indicator data such as geological exploration data, meteorological monitoring data, urban status data, and planning management data, as well as the actual development suitability level of each project, for example, divided into "high suitability", "medium suitability", and "low suitability". First, using the standardized indicator data as input features and the corresponding development suitability level as the prediction target, we use the random forest algorithm to supervise learning on the first training sample set composed of these 1000 project data, and initially train a random forest model. Further, in order to improve the performance of the random forest model, we optimize the hyperparameters of the random forest model. For example, through grid search, we try the number of trees between [50, 100, 150, 200], the maximum depth between [5, 10, 15, 20], and the minimum number of samples in the leaf nodes between [1, 2, 3, ...]. All combinations between 4]; for each combination, the random forest model is trained and evaluated on the validation set, and the hyperparameter combination that maximizes the accuracy of historical case evaluation is selected to generate the final weight correction model; in actual evaluation, the objective value weight of each primary indicator is first calculated using the entropy weight method, and the subjective experience weight is obtained using the analytic hierarchy process. Then, the two are weighted and summed at a ratio of 60% objective and 40% subjective to form a basic weight combination; further, this basic weight combination is input into the optimized weight correction model, which outputs a set of preliminary correction coefficients; to verify the reliability of the preliminary correction coefficients, five independent historical projects that did not participate in model training are selected as target historical cases, and the preliminary correction coefficients are applied to the selected target historical cases for evaluation. If the matching degree between the evaluation results and the actual suitability level of these historical projects reaches more than 90%, the verification can be considered successful, and this set of preliminary correction coefficients is used as the final dynamic correction coefficients.

[0074] The above technical solutions clarify the training, optimization, and verification process of the weight correction model, ensuring the accuracy and reliability of the generated dynamic correction coefficients. This enables the first-level evaluation weights of each first-level indicator in the multi-dimensional evaluation index to more accurately reflect the actual situation, thereby effectively improving the overall reliability and scientific nature of the suitability assessment of underground space resource development in coastal cities and avoiding evaluation deviations caused by improper weight settings.

[0075] In one embodiment, step S40 includes: S41: Obtain planning standard information and engineering specification information for the target area, and set grade classification thresholds for each secondary indicator in the multidimensional evaluation index based on the planning standard information and engineering specification information; In this embodiment, planning standard information may include documents such as urban master plans, underground space special plans, and land use plans. These documents typically specify the development intensity, functional positioning, and environmental protection requirements for different areas. Engineering specification information may include building foundation design requirements, geotechnical engineering investigation requirements, and underground engineering construction requirements. These requirements have specific technical indicators in terms of geological conditions, hydrological conditions, and structural safety. By collecting and interpreting planning standard information and engineering specification information, preliminary grade classification thresholds can be set for each secondary indicator in the multidimensional evaluation indicators. For example, specific numerical limits can be set for indicators such as groundwater level depth, soil bearing capacity, and geological disaster risk level to distinguish different suitability levels. Another approach is to refine and supplement the planning standards and engineering specifications by combining expert experience and historical project data to form a more comprehensive and practical initial grade classification threshold system.

[0076] S42: Obtain standardized indicator data and development suitability levels of historical engineering cases, and use the obtained standardized indicator data and development suitability levels as training features and training labels respectively to construct a second training sample set; In this embodiment, historical engineering cases refer to underground space development projects that have been completed in the target area or similar areas. These projects have clear development suitability assessment results or actual development effects. Standardized indicator data refers to the various secondary indicator data used in the evaluation of historical cases, such as soil layer parameters, hydrological monitoring data, and records of adverse geological phenomena in geological survey reports. Development suitability level refers to the final suitability level of historical cases, such as "very suitable", "suitable", "generally suitable", "unsuitable", etc. By using standardized indicator data as training features and development suitability level as training labels, a sample set containing the correspondence between input and output can be constructed for machine learning models to learn and generalize. In addition, data augmentation techniques, such as slightly perturbing or combining existing historical case data, can be used to expand the sample set size and improve the model's generalization ability.

[0077] S43: Train the logistic regression model using the second training sample set to generate a hierarchical threshold optimization model; In this embodiment, the logistic regression model is a generalized linear model commonly used in classification problems. It can predict the probability of an event occurring based on input features. In this embodiment, it is used to learn the nonlinear relationship between various indicator data in historical cases and the final development suitability level. By inputting the second training sample set into the logistic regression model, the model iteratively learns based on the training features and training labels, adjusting its internal parameters to minimize the error between the predicted level and the actual level. After training, the logistic regression model can predict its suitability level based on new indicator data and indirectly reflect the optimal grading threshold.

[0078] S44: Obtain the current standardized indicator data of the target area, and input the level classification threshold and the current standardized indicator data into the level threshold optimization model to obtain the optimized level threshold; In this embodiment, the current standardized indicator data refers to the secondary indicator data of the target area to be evaluated, which are acquired in real time and standardized. The grading threshold is the initial threshold set earlier according to planning standards and engineering specifications. The above information is input into the pre-trained grading threshold optimization model. The grading threshold optimization model will comprehensively consider the actual situation of the current area and historical experience to adjust and optimize the initially set grading threshold. For example, the grading threshold optimization model may fine-tune the suitability boundary of a certain indicator according to the special geological conditions or urban development needs of the current area to make it more in line with the actual situation. Another implementation method is that the grading threshold optimization model can directly output the optimal grading threshold for each indicator of the current area without relying on the preset initial threshold as input, thereby achieving a more thorough dynamic adjustment.

[0079] S45: Identify the target historical cases and apply the optimized grading threshold to the target historical cases for evaluation. If the evaluation result is satisfactory, output the optimized grading threshold as the grading threshold.

[0080] In this embodiment, the target historical case refers to one or more typical historical projects with clear evaluation results and actual development effects, used to verify the effectiveness of the optimized threshold. The threshold obtained by the hierarchical threshold optimization model is applied to the target historical case for re-evaluation, and its evaluation result is compared with the actual suitability level of the target historical case. If the evaluation result is highly consistent with the actual situation, or meets the preset evaluation criteria such as accuracy and recall, the optimized hierarchical threshold is considered reliable and effective, and can be formally adopted as the hierarchical threshold for the current evaluation task. If the evaluation result fails, it may be necessary to re-examine the model training process, sample set quality, or initial threshold setting, and perform iterative optimization.

[0081] For example, as a specific implementation method, when conducting a suitability assessment of a newly planned underground commercial district in a coastal city, the following steps can be taken: First, the threshold values ​​for secondary indicators such as foundation bearing capacity, groundwater level depth, soil liquefaction index, and geological structure stability can be extracted from documents such as the city's underground space planning, geological disaster prevention planning, and building foundation design requirements. For instance, a foundation bearing capacity greater than 250 kPa is defined as "highly suitable," 150-250 kPa as "medium suitable," and less than 150 kPa as "low suitable." Further, data from the city and surrounding areas can be collected... Data on underground engineering projects completed within the past decade, including detailed geological survey reports, hydrological monitoring records, construction logs, and suitability ratings in the final project acceptance reports; for example, a subway station project with a foundation bearing capacity of 280 kPa, a groundwater level depth of 15 m, and a soil liquefaction index of 0.1 was ultimately rated as "very suitable"; after standardizing the collected data, a second training sample set was constructed, and this second training sample set was used to train the logistic regression model; during the training process, the logistic regression model learns how to determine the suitability level based on the foundation bearing capacity, groundwater level, and other factors. The combination of numerical values ​​for indicators such as groundwater level and depth is used to predict the suitability level of underground space development. For example, a logistic regression model may find that when the groundwater level exceeds a certain value, even if the foundation bearing capacity is high, the suitability level will significantly decrease. When assessing a new commercial district, the latest geological survey data and hydrological monitoring data for the area are obtained to form current standardized indicator data. The obtained data, along with the previously set initial level classification thresholds, are input into a trained logistic regression model. The logistic regression model will fine-tune the initial thresholds based on the actual situation of the current area. For example, if the groundwater level in the current area fluctuates significantly, the logistic regression model may adjust the "high suitability" threshold for groundwater level depth from 15m to 18m to reflect the actual engineering risks. Finally, an underground parking project in the city with similar geological conditions to the commercial district to be assessed and which has been successfully operating for many years can be selected as a target historical case. The optimized classification thresholds are applied to the historical data of the parking project for evaluation. If the evaluation results are consistent with the actual suitability level of the parking project, the optimized classification thresholds can be considered effective and can be used for the final underground space development score calculation.

[0082] Through the above technical solutions, this application can effectively solve the problems of rigid grading threshold settings and difficulty in adapting to regional differences and dynamic changes in traditional underground space development suitability assessments. Specifically, by introducing planning standard information and engineering specification information as the initial setting basis, the compliance and fundamental nature of the assessment can be ensured. Furthermore, by using historical engineering case data to train a logistic regression model, the grading thresholds can learn from practical experience, achieving a deep understanding of complex geological environments and urban development needs. In actual assessments, the standardized data of the current region, along with the initial thresholds, can be input into the optimization model to generate highly targeted optimized grading thresholds. Finally, through verification with target historical cases, the reliability and accuracy of the optimized thresholds can be ensured, making the assessment results closer to reality and improving the scientific rigor and practicality of the assessment.

[0083] In one embodiment, step S43 includes: S431: Using the standardized index data and the level division threshold in the second training sample set as input features, and the development suitability level corresponding to the standardized index data in the second training sample set as the prediction target, several sample pairs are formed. In this embodiment, the second training sample set includes standardized indicator data of historical engineering cases and their corresponding development suitability levels. The standardized indicator data is combined with pre-set level classification thresholds as input features of the logistic regression model, which can reflect various factors affecting the development suitability of underground space and their quantitative boundaries. At the same time, the development suitability level of historical cases is used as the prediction target, enabling the logistic regression model to learn the mapping relationship between the input features and the final suitability level. The sample pairs constructed in this way can provide necessary data support for the learning and generalization of the logistic regression model, ensuring that the logistic regression model can learn from historical experience how to judge the suitability level based on specific indicators and thresholds. For example, the input features may include standardized values ​​of geological conditions, hydrological conditions, etc. of a historical project, as well as level classification thresholds set for these conditions, while the prediction target is the actual development suitability level of the historical project.

[0084] S432: Define the learning task of the logistic regression model, the learning task specifically including: generating a predicted development suitability level for sample pairs based on input features; In this embodiment, the definition of the learning task is crucial in guiding the model training direction. It specifies what the logistic regression model needs to learn from the input data and how to apply the learned knowledge to new data. Specifically, the goal of the logistic regression model is to receive input features consisting of standardized indicator data and grading thresholds, and output a predicted value based on the input features. This predicted value represents the suitability level for underground space development. The output predicted value can be a discrete classification result or a continuous probability value, representing the probability of belonging to a certain suitability level. For example, the learning task can be defined as follows: given geological, hydrological, and other indicator data of a project and its corresponding grading thresholds, the logistic regression model needs to predict the probability that the project belongs to a suitable development level.

[0085] S433: Using the objective function of minimizing the cross-entropy loss between the predicted development suitability level and the corresponding development suitability level in the sample pair, the logistic regression model is iteratively trained using the gradient descent algorithm to generate a hierarchical threshold optimization model.

[0086] In this embodiment, the objective function is a quantitative indicator that measures the difference between the model's predicted results and the actual results. Minimizing this objective function is the driving force for the logistic regression model's learning. The cross-entropy loss function is particularly suitable for classification problems, as it effectively measures the difference between the predicted probability distribution and the true label distribution. By minimizing the cross-entropy loss, the logistic regression model can continuously adjust its internal parameters, making the predicted results increasingly closer to the true values. Gradient descent is a commonly used optimization algorithm. It calculates the gradient of the loss function with respect to the logistic regression model parameters and updates the parameters in the opposite direction of the gradient, thereby gradually reducing the value of the loss function. This allows the logistic regression model to learn complex nonlinear relationships from a large number of sample pairs, ultimately generating a hierarchical threshold optimization model that can accurately predict development suitability levels. For example, in each iteration, the logistic regression model calculates the predicted value based on the current parameters, and then fine-tunes the parameters using gradient descent based on the cross-entropy loss between the predicted and actual values ​​until the loss function converges or reaches the preset training epochs.

[0087] For example, as a specific implementation method, when constructing sample pairs, a historical engineering case can be selected from the second training sample set. Its standardized indicator data may include specific values ​​such as geological bearing capacity, groundwater level depth, and soil corrosivity, as well as grade classification thresholds set for these indicators. For example, geological bearing capacity greater than 200 kPa is high, 100-200 kPa is medium, and less than 100 kPa is low, and the actual development suitability level of this historical engineering case is suitable. The selected data constitutes a sample pair, where the standardized indicator data and grade classification thresholds serve as input features, and the development suitability level serves as the prediction target. When defining the learning task of the logistic regression model, it can be set as a multi-classification task, that is, based on the input standardized indicators... Based on data and grading thresholds, the system predicts the development suitability level of the project. During the logistic regression model training phase, the Adam optimizer can be used as a specific implementation of the gradient descent algorithm, with a learning rate of 0.01 and a batch size of 32. In each iteration, the logistic regression model calculates the cross-entropy loss between the current prediction and the true suitability level. Then, the Adam optimizer automatically adjusts the weights and bias parameters of the logistic regression model based on the cross-entropy loss value and the learning rate. The training process can continue until the loss function converges on the validation set or reaches the preset maximum training epochs, such as 100 epochs. Finally, the trained and optimized logistic regression model is used as a grading threshold optimization model for suitability assessment.

[0088] The above technical solution uses standardized indicator data and grading thresholds as input features, and takes the actual development suitability level as the prediction target to construct high-quality training samples. This allows the logistic regression model to fully learn the key factors affecting the suitability level and their quantification boundaries. By clearly defining the learning task, the accuracy of the model training direction is ensured. At the same time, iterative training using the cross-entropy loss function and gradient descent algorithm enables the logistic regression model to efficiently learn and optimize parameters from historical data, thereby generating an accurate grading threshold optimization model. This improves the scientific validity of grading threshold determination and the reliability of evaluation results.

[0089] In one embodiment, step S50 includes: S51: Compare the secondary indicators corresponding to each sampling point in the standardized dataset with the grading threshold, and calculate the single-factor score of each secondary indicator through the preset membership function; In this embodiment, by comparing with the grading threshold and using a membership function, the standardization and fuzzification of secondary indicators with different properties and dimensions can be achieved, enabling them to participate in the comprehensive evaluation. For example, the trapezoidal membership function or Gaussian membership function in fuzzy mathematics can be used to calculate the membership degree of the indicator value to the corresponding level based on the position of the indicator value in different grading threshold intervals, thereby obtaining a single-factor score between 0 and 1. Alternatively, a piecewise linear function or a step function can be used as the membership function, and by setting different threshold intervals, the indicator value can be mapped to discrete or continuous scores.

[0090] S52: Based on the combined weighting strategy, determine the secondary evaluation weights of each secondary indicator in the multidimensional evaluation index, and aggregate them by weighted summation according to the single-factor scores to calculate the multi-factor scores of engineering geological conditions, hydrogeological conditions, adverse geological indicators and urban development demand indicators respectively. In this embodiment, the combined weighting strategy can combine subjective weighting methods, such as expert scoring or analytic hierarchy process, and objective weighting methods, such as entropy weighting or CRITIC method, to obtain the final secondary evaluation weights through linear or multiplicative combination. Alternatively, historical case data can be used to learn the relationship between secondary indicators and multifactor scores through machine learning methods, such as regression models or neural network models, thereby determining the secondary evaluation weights.

[0091] S53: Calculate the comprehensive suitability score of each geographic grid in the target area based on the preset scoring weight ratio of each primary indicator and its corresponding multi-factor score. In this embodiment, the preset scoring weight ratio can be determined by expert experience, strategy guidance, or historical data analysis. For example, based on the city's development direction and strategy, the city's development needs indicators can be given higher weights. During calculation, the multi-factor score of each primary indicator can be directly multiplied by its preset scoring weight ratio, and then the results of all primary indicators can be summed to obtain the comprehensive suitability score.

[0092] S54: Based on the comprehensive suitability score, a continuous spatial distribution score map is generated using spatial interpolation to obtain the underground space development score of the target area.

[0093] In this embodiment, spatial interpolation can compensate for the lack of data sampling points and provide continuous evaluation information for the entire region. Among them, the spatial interpolation method can be Kriging interpolation, which takes into account the autocorrelation of spatial data and can provide the best unbiased estimate. Alternatively, inverse distance weighting or spline function interpolation can be used to estimate the value of unknown points by weighting the known points or fitting a smooth surface.

[0094] Specifically, the proposed solution first compares the secondary indicators corresponding to each sampling point in the standardized dataset with preset grading thresholds, and then uses a membership function to transform them into single-factor scores with uniform dimensions, thereby achieving standardized processing of heterogeneous indicators. Further, based on a combined weighting strategy, it determines the secondary evaluation weights for each secondary indicator, and combines these with the single-factor scores for weighted summation, aggregating the scores to obtain multi-factor scores for each primary indicator, such as engineering geological conditions, hydrogeological conditions, adverse geological indicators, and urban development demand indicators. This ensures that the evaluation process fully considers the independent contribution of each secondary indicator and its relative importance in the primary indicators. On this basis, the multi-factor scores of each primary indicator are weighted and calculated with their preset weight proportions to obtain the comprehensive suitability score for each geographic grid within the target area. To transform the discrete grid scores into intuitive and continuous evaluation results, this solution further employs spatial interpolation to generate a continuous spatial distribution score map of the entire target area, ultimately obtaining the underground space development score.

[0095] Through the above technical solutions, this application can achieve a multi-level assessment of the suitability of underground space resource development in coastal cities. Specifically, by converting each secondary indicator into a unified single-factor score and combining it with a combined weighting strategy for multi-factor aggregation, the problem of difficulty in directly comparing and synthesizing different types of indicators can be effectively solved. At the same time, by calculating the comprehensive suitability score of each geographic grid and using spatial interpolation to generate a continuous spatial distribution score map, the spatial accuracy and intuitiveness of the assessment results can be improved, enabling the assessment results to clearly reflect the suitability differences of different locations within the target area, thereby improving the scientificity and effectiveness of underground space resource development decisions.

[0096] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0097] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method for assessing the suitability of underground space resource development in coastal cities based on multi-source data, characterized in that, Including the following steps: The target area is identified and a multidimensional evaluation index is established, which includes several primary indicators. The primary indicators include engineering geological conditions, hydrogeological conditions, adverse geological indicators, and urban development demand indicators. The hydrogeological conditions and adverse geological indicators respectively integrate a first dynamic evaluation factor and a second dynamic evaluation factor. The system acquires multi-source datasets of the target area in real time and performs spatiotemporal alignment processing on the multi-source datasets to generate a standardized dataset. The standardized dataset includes geological exploration data, meteorological monitoring data, urban status data, and planning and management data. Based on a standardized dataset, the first-level evaluation weights of each first-level indicator in the multidimensional evaluation index are determined by a pre-set combination weighting strategy, and the grading thresholds of each first-level indicator in the multidimensional evaluation index are determined by a pre-trained grading threshold optimization model. The standardized dataset is weighted and fused according to the primary evaluation weight and the hierarchical threshold to generate an underground space development score for the target area. The suitability level of the target area is determined based on the underground space development score, and a development decision assessment report corresponding to the suitability level is generated.

2. The method for assessing the suitability of underground space resource development in coastal cities based on multi-source data as described in claim 1, characterized in that: The step of determining the target area and establishing a multidimensional evaluation index including several primary indicators, wherein the primary indicators include engineering geological conditions, hydrogeological conditions, adverse geological indicators, and urban development demand indicators, wherein the hydrogeological conditions and adverse geological indicators respectively integrate a first dynamic evaluation factor and a second dynamic evaluation factor, includes: Engineering geological conditions, hydrogeological conditions, adverse geological indicators, and urban development demand indicators are used as primary indicators in the multi-dimensional evaluation indicators. Obtain the technical standards for the primary indicators and the development requirements for the target area, and decompose each primary indicator into its corresponding set of secondary indicators based on the technical standards and development requirements of the target area. Obtain the first historical dataset of the target area, and based on the first correlation model and the first historical dataset, screen out typhoon impact parameters from the secondary indicators of hydrogeological conditions, and integrate them into the first dynamic evaluation factor. The second historical dataset of the target area is obtained, and based on the second correlation model and the second historical dataset, typhoon disturbance parameters are screened from the secondary index set of adverse geological indicators and integrated into the second dynamic evaluation factor.

3. The method for assessing the suitability of underground space resource development in coastal cities based on multi-source data as described in claim 2, characterized in that: The step of obtaining the first historical dataset of the target area, and based on the first correlation model and the first historical dataset, screening typhoon impact parameters from the secondary index set of hydrogeological conditions, and integrating them into the first dynamic evaluation factor, includes: Historical typhoon information of the target area and its corresponding first time series data are obtained to form a first historical dataset. The first time series data includes a secondary index set of engineering geological conditions, a secondary index set of hydrogeological conditions, and typhoon characteristic parameters. Using typhoon characteristic parameters and secondary index sets of engineering geological conditions as input, and secondary index sets of hydrogeological conditions as prediction targets, the first correlation model is generated by training the first historical dataset through a gradient boosting tree model. The preset typhoon conditions and their corresponding engineering geological conditions are input into the first correlation model to calculate the predicted change rate of the secondary index set of engineering geological conditions and the secondary index set of hydrogeological conditions under the preset typhoon conditions. Secondary indicators whose predicted rate of change exceeds a preset sensitivity threshold are identified as typhoon impact parameters, and these typhoon impact parameters are integrated into the first dynamic evaluation factor.

4. The method for assessing the suitability of underground space resource development in coastal cities based on multi-source data as described in claim 2, characterized in that: The step of obtaining the second historical dataset of the target area, and based on the second correlation model and the second historical dataset, screening typhoon disturbance parameters from the secondary index set of adverse geological indicators, and integrating them into the second dynamic evaluation factor, includes: Historical typhoon information and its corresponding second time series data for the target area are obtained to form a second historical dataset. The second time series data includes a set of secondary indicators of adverse geological indicators, a set of secondary indicators of hydrogeological conditions, and typhoon characteristic parameters. Using the secondary index set of typhoon characteristic parameters and hydrogeological conditions as input, and the secondary index set of adverse geological indicators as the prediction target, a second correlation model is generated by training the second historical dataset through a logistic regression model. The preset typhoon conditions and their corresponding hydrogeological conditions are input into the second correlation model to calculate the predicted occurrence rate or risk change rate of the secondary index set of adverse geological indicators under the preset typhoon conditions. Secondary indicators that indicate the predicted incidence rate or risk change rate exceeds a preset risk threshold are defined as typhoon disturbance parameters, and these typhoon disturbance parameters are integrated into a second dynamic assessment factor.

5. The method for assessing the suitability of underground space resource development in coastal cities based on multi-source data as described in claim 1, characterized in that: The steps of acquiring multi-source datasets of the target area in real time, performing spatiotemporal alignment processing on the multi-source datasets to generate a standardized dataset, wherein the standardized dataset includes geological exploration data, meteorological monitoring data, urban status data, and planning management data, include: Deep learning OCR is used to parse unstructured data from multi-source datasets, extract effective information, and transform it into semi-structured data. Based on a pre-defined unified spatial reference coordinate system, coordinate transformation and mapping matching are performed on spatial data in multi-source datasets to achieve spatial dimension alignment. Based on the time-series data in the multi-source dataset with a preset time granularity, a unified timestamp is labeled for the non-time-series data, and the gaps in the data sequence are filled by interpolation to achieve time dimension alignment. The 3σ criterion is used to verify the multi-source dataset after spatial and temporal alignment, identify outlier data points and correct them, and obtain the outlier processing results. Based on the outlier handling results, invalid data with deviations exceeding a preset deviation threshold in the multi-source dataset are excluded, and the remaining data after exclusion are processed uniformly to generate a standardized dataset.

6. The method for assessing the suitability of underground space resource development in coastal cities based on multi-source data as described in claim 1, characterized in that: The steps of determining the first-level evaluation weights of each first-level indicator in the multidimensional evaluation index based on a standardized dataset and a preset combined weighting strategy, and determining the grading thresholds of each first-level indicator in the multidimensional evaluation index using a pre-trained grading threshold optimization model, include: The objective value weights of each primary indicator were calculated using the entropy weight method, and the subjective experience weights of each primary indicator were obtained through a pre-defined hierarchical analysis strategy. Obtain standardized indicator data and development suitability levels from historical engineering cases, and use the obtained standardized indicator data and development suitability levels as training features and training labels respectively to construct the first training sample set; The random forest model is trained using the first training sample set to generate a weight correction model. The linear combination of objective value weights and subjective experience weights is then input into the weight correction model to obtain dynamic correction coefficients. The linear combination of objective value weights and subjective experience weights is dynamically adjusted based on the dynamic correction coefficient, and the primary evaluation weights of each primary indicator are output.

7. The method for assessing the suitability of underground space resource development in coastal cities based on multi-source data as described in claim 6, characterized in that: The steps of training the random forest model using the first training sample set to generate a weight correction model, and inputting the linear combination of objective value weights and subjective experience weights into the weight correction model to obtain dynamic correction coefficients include: Using standardized index data from the first training sample set as input features and the development suitability level corresponding to the standardized index data as the prediction target, the first training sample set is subjected to supervised learning through the random forest algorithm. The hyperparameters of the random forest model, namely the number of trees, maximum depth, and minimum number of samples per leaf node, are optimized by grid search. The evaluation accuracy of historical cases is used as the objective function to train and generate a weight correction model. The weights of objective value and subjective experience are weighted and summed to form a basic weight combination. This basic weight combination is then input into the weight correction model to generate preliminary correction coefficients for the weights of each primary indicator. Identify target historical cases and apply the preliminary correction coefficients to them for verification. If the verification results are successful, output the preliminary correction coefficients as dynamic correction coefficients.

8. The method for assessing the suitability of underground space resource development in coastal cities based on multi-source data as described in claim 2, characterized in that: The steps of determining the first-level evaluation weights of each first-level indicator in the multidimensional evaluation index based on a standardized dataset and a preset combined weighting strategy, and determining the grading thresholds of each first-level indicator in the multidimensional evaluation index using a pre-trained grading threshold optimization model, include: Obtain planning standards and engineering specifications information for the target area, and set grade classification thresholds for each secondary indicator in the multidimensional evaluation index based on the planning standards and engineering specifications information; Obtain standardized indicator data and development suitability levels from historical engineering cases, and use the obtained standardized indicator data and development suitability levels as training features and training labels respectively to construct a second training sample set; The logistic regression model is trained using the second training sample set to generate a hierarchical threshold optimization model. Obtain the current standardized indicator data of the target area, and input the level classification threshold and the current standardized indicator data into the level threshold optimization model to obtain the optimized level threshold; Identify target historical cases and apply the optimized grading threshold to evaluate them. If the evaluation result is satisfactory, output the optimized grading threshold as the grading threshold.

9. The method for assessing the suitability of underground space resource development in coastal cities based on multi-source data as described in claim 8, characterized in that: The step of training the logistic regression model using the second training sample set to generate a hierarchical threshold optimization model includes: Using the standardized index data and the level division threshold in the second training sample set as input features, and the development suitability level corresponding to the standardized index data in the second training sample set as the prediction target, several sample pairs are formed. Define the learning task of the logistic regression model, which specifically includes: generating a predicted development suitability level for sample pairs based on input features; Using the minimization of the cross-entropy loss between the predicted development suitability level and the corresponding development suitability level in the sample pair as the objective function, the logistic regression model is iteratively trained using the gradient descent algorithm to generate a hierarchical threshold optimization model.

10. The method for assessing the suitability of underground space resource development in coastal cities based on multi-source data as described in claim 2, characterized in that: The step of generating an underground space development score for the target area by weighted fusion calculation of the standardized dataset based on the primary evaluation weight and the hierarchical threshold includes: The secondary indicators corresponding to each sampling point in the standardized dataset are compared with the grading threshold, and the single-factor scores of each secondary indicator are calculated through the preset membership function. The secondary evaluation weights of each secondary indicator in the multidimensional evaluation index are determined based on the combined weighting strategy. Based on the single-factor scores, the weighted summation is used to aggregate the scores and calculate the multi-factor scores of engineering geological conditions, hydrogeological conditions, adverse geological indicators and urban development demand indicators respectively. Based on the preset scoring weights of each primary indicator and their corresponding multi-factor scores, the comprehensive suitability score of each geographic grid in the target area is calculated. Based on the comprehensive suitability score, a continuous spatial distribution score map is generated using spatial interpolation to obtain the underground space development score of the target area.

Images