Health risk prediction and intervention assessment system based on urban heat island stress identification

Abstract

This invention discloses a health risk prediction and intervention assessment system based on urban heat island stress identification. The system includes a data collection module for acquiring health and disease information, urban heat island exposure data, and proteomics and metabolomics data; a data processing module for automated preprocessing and data standardization of multi-source heterogeneous data, constructing a unified research and analysis database; an artificial intelligence analysis module for screening key molecular features across multiple omics, constructing health risk prediction models, and adaptively optimizing parameters; a scenario simulation and feedback module for simulating changes in urban heat island intensity under different mitigation scenarios, generating multi-dimensional prediction and feedback data; and a decision support module for generating individualized health management recommendations and group intervention priority plans. This invention addresses current problems in urban heat island health management, such as insufficient exposure assessment accuracy, lack of multi-omics mechanism identification, limited risk prediction capabilities, and the inability to quantitatively assess intervention effects.

Description

Technical Field

[0001] This invention relates to a health risk prediction and intervention assessment system, and more particularly to a health risk prediction and intervention assessment system based on urban heat island stress identification. Background Technology

[0002] The urban heat island (UHI) phenomenon leads to significant spatial temperature differences within cities. Factors such as urban morphology, land cover, building materials, transportation, and green space distribution result in high heterogeneity in heat exposure among different neighborhoods and individuals within the same city. Recent systematic reviews and urban heat island mapping studies show that relying solely on city-level or station-level average meteorological data cannot reflect the microscale exposure differences among residents. This exposure heterogeneity is key to explaining the spatial distribution of heat-related diseases and health inequalities within cities. Therefore, high-resolution characterization of individualized and spatially refined heat exposure is fundamental to accurately identifying heat stress-related risks and developing differentiated intervention strategies. Regarding the quantification of heat exposure, most current studies rely on satellite remote sensing of land surface temperature, meteorological station temperatures, or high-resolution air temperature grid data constructed through empirical extrapolation. While these methods can provide information on regional-scale temperature changes, they are limited by factors such as spatial resolution, temporal update frequency, underlying surface heterogeneity, and local microenvironmental shading, making it difficult to accurately reflect the true individual heat exposure levels of residents. In addition, existing methods have limited consideration of the dynamic interaction of urban form elements such as urban building form, vegetation cover, ventilation corridors, and transportation networks, resulting in a systematic bias in the capture of "real perceived thermal exposure".

[0003] Heat exposure affects the body through multiple physiological and molecular pathways, such as inducing inflammatory responses, oxidative stress, metabolic disorders, and altered vascular function. In recent years, "external exposure-intrinsic biological effects" studies have increasingly incorporated proteomics, metabolomics, and other multi-omics data into environmental epidemiology research to reveal the biological changes caused by environmental exposure from a molecular perspective. Metabolomics and proteomics are highly sensitive in reflecting short- and medium-term physiological responses, providing important evidence for determining biomarkers of heat stress, identifying mediating pathways, and validating intervention effects. Meanwhile, the integration of systemic exomics, transcriptomics, proteomics, and other multi-omics approaches has been considered a key pathway for advancing precision environmental health research. However, the integration of multi-omics in environmental research is still in a stage of rapid methodological development and faces challenges related to sample size, batch effects, and high-dimensional statistical synergies, requiring further improvement in data standardization, model robustness, and interpretability.

[0004] In terms of data analysis methods and modeling, artificial intelligence (especially deep learning) and ensemble machine learning provide powerful tools for the fusion of multimodal and multi-scale data. Related research has demonstrated that multimodal machine learning can improve predictive performance based on omics and environmental data, and can identify complex nonlinear relationships and interactions in large sample cohorts. However, the application of AI models in environmental health still faces several key challenges: first, insufficient interpretability and biological consistency of the models; second, susceptibility to data bias and confounding; and third, the generalizability of multi-omics fusion strategies (such as feature-level fusion, model-level fusion, or network-level fusion) across different tasks remains unclear. Combining causal inference and mediation analysis methods can enhance the causal orientation of model findings, which is particularly important for transforming predictive results into actionable intervention recommendations. While there is technical capability to integrate exposure, omics, and health outcomes into individualized risk scores, a standardized closed-loop system from technology to application is still lacking.

[0005] In terms of health risk prediction, traditional models typically rely on meteorological high-temperature indicators, past medical history, or demographic characteristics for risk assessment. However, these models cannot utilize molecular-level signals or reflect individual differences in biological sensitivity to heat exposure, making risk prediction dependent on macroscopic characteristics and hindering mechanism-driven prediction and early warning. Similarly, for urban heat adaptation interventions (such as greening, shading, ventilation design, and cooling modifications), existing assessments are largely based on environmental improvement indicators (such as temperature reduction and improved thermal comfort), lacking means to quantify the intervention effects at the individual molecular response level. The lack of omics evidence makes it difficult to verify the effectiveness of interventions from a mechanistic perspective and to construct a feedback-based, iteratively optimized framework for heat exposure health management.

[0006] As can be seen from the above, existing technologies have systemic shortcomings in the following aspects: First, exposure quantification is still insufficient in terms of spatial-temporal resolution and coupling with individual behavior; second, evidence on the molecular mechanisms of heat stress is scattered and lacks multi-omics integrated research; third, although AI / machine learning can improve predictive performance, it still lacks interpretability and causal inference; fourth, the evaluation of intervention effects mainly focuses on macro-health outcomes and behavioral indicators, lacking quantitative evaluation based on molecular markers and closed-loop feedback mechanisms. These shortcomings directly lead to an "information gap" in the chain of heat health management, from population prediction to individualized intervention and then to policy implementation. Therefore, it is necessary to construct an integrated system that simultaneously possesses the capabilities of heat island stress response identification, health risk prediction, and intervention effect evaluation. Summary of the Invention

[0007] The technical problem to be solved by the present invention is to provide a health risk prediction and intervention assessment system based on urban heat island stress identification, which can solve the core technical problems in current urban heat island health management, such as insufficient exposure assessment accuracy, lack of multi-omics mechanism identification, limited risk prediction ability, and inability to quantitatively assess intervention effects.

[0008] To address the aforementioned technical problems, this invention provides a health risk prediction and intervention assessment system based on urban heat island stress identification. The system includes a data collection module for collecting baseline information of research subjects, continuously acquiring health and disease information, urban heat island exposure data, and proteomics and metabolomics data during follow-up, and supporting automatic access to multi-source environmental data from satellite remote sensing, meteorological station networks, and urban spatial information platforms; a data processing module for performing automated preprocessing and data standardization on the collected multi-source heterogeneous data, and based on intelligent data fusion algorithms, performing spatiotemporal matching and multimodal alignment of environmental exposure information, molecular-level information, and individual health information to construct a structurally unified research and analysis database; and an artificial intelligence analysis module for embedding machine learning and deep learning algorithm frameworks to automatically perform feature engineering and transformation on the exposure-molecular-health information in the database. The module performs quantitative selection and model training, completing the screening of key molecular features in multi-omics, constructing a health risk prediction model, and adaptively optimizing parameters. It continuously updates model weights based on an algorithm tuning mechanism, enabling dynamic learning and iterative optimization of model performance. The scenario simulation and feedback module constructs a virtual intervention environment based on the AI ​​prediction model, simulating changes in urban heat island intensity under different mitigation scenarios. It outputs trends in individual risk levels, changes in molecular stress responses, and changes in disease probability under various assumptions, forming multi-dimensional prediction feedback data. The decision support module intelligently analyzes and sorts the above prediction feedback data, combining risk value changes, absolute benefit indicators, and population distribution structure to generate individualized health management recommendations and group intervention priority plans. It also outputs policy evaluation reports, risk classification lists, or intervention benefit interpretation documents through a visual interface.

[0009] Furthermore, the data collection module integrates satellite remote sensing, meteorological monitoring, and urban morphology databases based on baseline information and spatial location data to construct an urban heat island exposure model covering the study area and generate corresponding individualized heat exposure indicators for each research object.

[0010] Furthermore, the data collection module specifically includes: First, acquiring multi-source remote sensing observation data, including medium- and high-resolution land surface temperature, normalized difference vegetation index, land surface albedo, land use and land cover data, and performing quality control and spatiotemporal registration on these data; then, identifying urban areas using land surface characteristic indicators, and constructing a stable urban boundary mask by combining administrative boundaries and light and night remote sensing data; after obtaining the urban area, calculating the background temperature using the urban-suburban comparison method; finally, comparing the land surface temperature of each grid point within the urban area with its corresponding background temperature, obtaining the heat island intensity value by calculating grid by grid, and generating an individualized urban heat island exposure index for each research object based on its exposure location and activity time.

[0011] Furthermore, the formula for calculating the intensity value of the heat island is as follows: ； in, For urban areas in spatial location and time Surface or temperature value; The background temperature is the same as that of a reference station in the suburbs or rural areas during the same period.

[0012] Furthermore, the data processing module performs intelligent verification on data from different sources, including detecting time conflicts, duplicate records, abnormal fluctuation values, and missing fields, and automatically marks or removes data that does not meet quality standards; in terms of identity association, it integrates and associates information of the same research subject from different data sources through a unified identification key; in terms of spatiotemporal alignment, it ensures consistency between environmental exposure data and individual information in terms of time and space through a unified timestamp and spatial coordinate matching mechanism.

[0013] Furthermore, the data processing module specifically includes: performing consistency verification and missing value checks on the acquired baseline information of the research subjects; marking, correcting, or removing records with logical conflicts, incompleteness, or abnormal values; standardizing health outcome data by converting disease diagnosis data, death information, and follow-up records from multiple sources or different coding systems into a standard disease coding format, and labeling the follow-up status of the research subjects according to the event occurrence time; spatially and temporally aligning environmental exposure data by processing urban thermal environment data according to a preset spatial unit resolution and performing spatial matching based on the residential address information of the research subjects to form individual-level environmental exposure indicators; integrating baseline information, health and disease data, and matched environmental exposure data according to the individual identifiers of the research subjects to generate an analysis database containing individual characteristics, outcome events, and long-term environmental exposure levels; and finally, recording versions and storing the completed research analysis database in a structured manner.

[0014] Furthermore, the artificial intelligence analysis module performs standardization, missing value imputation, outlier identification, and variable recoding operations on proteomics and metabolomics data, converting the original multidimensional molecular signals into a feature expression space that can be learned by the model; and based on the integrated multi-omics data, urban heat island exposure level, and baseline covariates, a feature engineering module is constructed to uniformly map protein expression levels, metabolite intensity, heat island exposure value, and individual characteristics to the machine learning modeling space.

[0015] Furthermore, the artificial intelligence analysis module employs an Elastic Net regularized regression model to screen key molecular features significantly associated with urban heat island stress from high-dimensional proteomics and metabolomics data, achieving dimensionality reduction and variable screening at the molecular level. Its objective function is expressed as: ;in: For individuals At the molecular level, For the first A sample was exposed to the heat island or covariate. The value on, The regression coefficients to be estimated are: For regularization strength, Control the mixing ratio of Lasso regression and Ridge regression.

[0016] Furthermore, after obtaining the key molecular feature set, the artificial intelligence analysis module uses urban heat island exposure indicators, screened multi-omics molecular features, and individual baseline information as input variables; it constructs a health outcome prediction network with a survival analysis model at its core, and completes variable combination optimization and model parameter updates through machine learning mechanisms to achieve joint modeling of urban heat island exposure, multi-omics molecular features, and health risks; its risk function is expressed as: ; in :individual In time The instantaneous risk; Used as the benchmark risk function; Individual heat island exposure value; For the first A molecular signature associated with heat island exposure; For covariates; The corresponding regression coefficients obtained through maximum likelihood iterative optimization are trained by the artificial intelligence module.

[0017] Furthermore, the artificial intelligence analysis module introduces an absolute risk reduction index, the calculation formula of which is: ; in, The predicted risk of morbidity or mortality under a reference scenario; The decision support module predicts the risks under the intervention scenario; it ranks and optimizes each intervention plan based on the absolute risk reduction index and the overall risk change trend, and further converts them into the number of avoidable cases, the number of avoidable deaths, or the reduction in molecular damage.

[0018] Compared with existing technologies, the present invention has the following beneficial effects: The health risk prediction and intervention assessment system based on urban heat island stress identification provided by the present invention constructs an integrated system that simultaneously possesses the capabilities of heat island stress response identification, health risk prediction, and intervention effect assessment. It achieves a complete closed loop between "high-resolution exposure assessment - multi-omics mechanism analysis - interpretable AI risk prediction - molecularly validated intervention effect assessment", thereby solving the core technical problems in current urban heat island health management, such as insufficient exposure assessment accuracy, lack of multi-omics mechanism identification, limited risk prediction capabilities, and inability to quantitatively assess intervention effects. Attached Figure Description

[0019] Figure 1 This is a flowchart of the multi-omics fusion health risk prediction and intervention assessment based on artificial intelligence, as described in this invention. Figure 2 This is a diagram illustrating the architecture of the health risk prediction and intervention assessment system based on urban heat island stress identification, as presented in this invention. Figure 3 This is a diagram illustrating the automatic identification effect of molecular response signals under complex environmental exposure factors according to the present invention. Detailed Implementation

[0020] The present invention will now be further described with reference to the accompanying drawings and embodiments.

[0021] Before detailing the specific implementation methods of this application, to ensure consistency in understanding the specification, some terms and statistical methods involved in this application are clearly explained, including but not limited to: urban heat island intensity (UHI), which refers to the difference between the temperature of an urban grid (or point) estimated using remote sensing or ground meteorological data and the background reference temperature of a selected suburb, measured in degrees Celsius (°C); Cox proportional hazards regression model, a survival analysis model used to assess the relationship between the time of event occurrence and predictor variables; hazard ratio (HR), a relative risk indicator used to measure the ratio of the risk functions of individuals in different exposure groups; elastic net regression, used to screen key molecules significantly associated with heat exposure or biophenotypes from high-dimensional proteomic / metabolomic features; absolute risk reduction (ARR), used to measure the absolute difference in the probability of event occurrence between the control group and the intervention group; and concordance index (C-index), the proportion of individuals correctly ranked as high-risk or low-risk by the model among all comparable pairs.

[0022] Before describing the proposed method and system for urban heat island stress identification, health risk prediction, and intervention effect evaluation based on artificial intelligence and multi-omics fusion, this application first declares that this study strictly adheres to the principles of data privacy and ethical review. All data involving research subjects, including but not limited to baseline information, health and disease data, multi-omics data, and geospatial location data, must be obtained with the explicit consent of the research subjects and protected for privacy through methods such as identity de-identification and encrypted storage. The baseline information to be collected from research subjects includes demographic characteristics (age, gender, ethnicity, education level, etc.), socioeconomic factors (income, occupation category, type of work, etc.), lifestyle (smoking, drinking, physical activity, sleep, diet, etc.), past medical history (hypertension, diabetes, coronary heart disease, stroke, mental illness, etc.), and long-term medication use. The above information provides a structured foundation for subsequent construction of individualized urban heat island exposure models, analysis of molecular stress responses, and prediction of health risks.

[0023] The embodiments of the present invention will be further described below with reference to the accompanying drawings. In one aspect, the embodiments of the present invention provide a method for identifying urban heat island stress, predicting health risks, and evaluating intervention effects based on artificial intelligence and multi-omics fusion, referring to… Figure 1 This includes the following steps: Obtain baseline information of the research subjects; Based on baseline information, health and disease data, urban heat island exposure data, and multi-omics data were collected. Baseline information, health and disease data, exposure data, and multi-omics data are matched and integrated to establish a research and analysis database; Based on the analysis of the integrated database, predictive feedback data for different intervention or mitigation scenarios are generated. Based on the model feedback data, individualized or group-based intervention effect evaluation results are output.

[0024] The present invention discloses the following steps for obtaining baseline information of research subjects: Identify the research subjects; determine the included population through cohort design and regional health registration systems, and clarify the inclusion and exclusion criteria.

[0025] The identities of research subjects are anonymized to obtain data records that can be used for research. Baseline information will be collected, including: lifestyle information (smoking, alcohol consumption, physical activity, dietary habits, sleep, etc.); physical condition information (gender, age, BMI, blood pressure, history of chronic diseases, etc.); living environment information (residential address, floor, building type, green space exposure, etc.); and socioeconomic information (education level, income level, occupation type, etc.). This information will be used for subsequent exposure assessment, model adjustment, and results interpretation.

[0026] The present invention discloses a method for collecting health and disease data, urban heat island exposure data, and multi-omics data based on baseline information, including the following steps: Based on baseline information, lifestyle and physical condition information are analyzed, and health and disease data, including mortality and chronic non-communicable diseases, are obtained. The implementation details for measuring urban heat island exposure are constructed based on the remote sensing-meteorological fusion modeling technology proposed in existing research. This method uses high spatiotemporal resolution land surface temperature products as its core, and through steps such as urban boundary identification, background temperature estimation, rasterization interpolation, and exposure matching calculation, forms a continuous index that can characterize the true long-term heat exposure level of individuals. Its construction process includes the following key steps: First, acquiring multi-source remote sensing observation data, including medium-to-high resolution land surface temperature (LST), normalized difference vegetation index, surface albedo, land use, and land cover data, and performing quality control and spatiotemporal registration. By using diurnal LST data, the changes in the heat island at different time periods can be effectively captured, avoiding dependence on a single temporal hotspot. Subsequently, urban areas are identified using surface characteristic indicators (such as building density, vegetation cover, and bare land ratio), and a stable urban boundary mask is constructed by combining administrative boundaries and light and nighttime remote sensing data. After obtaining the urban area, the background temperature is calculated using an urban-suburban comparison method: temperature samples are taken from low-density built-up areas or natural surface areas outside the city, and spatial interpolation (such as spline interpolation, inverse distance weighting, or kriging) is used to construct the regional background temperature field. The background temperature represents the natural baseline thermal environment undisturbed by urban construction and serves as a reference for assessing urban warming signals. Subsequently, the LST of each grid point within the urban area is compared with its corresponding background temperature, and the heat island intensity value is calculated grid by grid. ； in, For urban areas in spatial location and time Surface or temperature value; The background temperature is used as a reference station in suburban or rural areas during the same period. This formula generates an individualized urban heat island exposure index for each study subject based on their exposure location and activity time. To reduce the impact of instantaneous fluctuations in weather conditions, the urban heat island data undergoes further meteorological standardization processing. This involves constructing a calibration model using meteorological parameters such as temperature, humidity, wind speed, radiation, and seasonal cycles during the same period, eliminating mean drift caused by large-scale weather systems. The resulting index is more stable and better reflects the structural warming caused by urban development characteristics. To construct a long-term exposure index, urban heat island values ​​from consecutive years are aggregated over time. For example, annual or multi-year data are weighted and averaged by day, week, or month. The rasterized urban heat island values ​​are spatially linked to individual residential addresses through geocoding, and the corresponding exposure value is extracted based on the residential location. For cases involving changes in residence, a multi-stage UHI can be calculated based on the time-weighted address change, ensuring that the obtained long-term exposure accurately reflects the actual thermal pressure of an individual's living environment at different times. The resulting urban heat island exposure index is continuous, spatially accurate, and temporally stable. It can comprehensively reflect the enhancing effect of factors such as urban surface structure, vegetation cover, energy consumption, and surface morphology on the local temperature field, providing a high-quality exposure measurement basis for subsequent research on the impact of environmental thermal stress on health, metabolomics, or proteomics pathways.

[0027] As an optional implementation, the urban heat island exposure measurement process in this invention can be flexibly adjusted according to the actual application scenario, including replacing remote sensing data sources, optimizing spatial resolution, and changing meteorological correction models. For areas with missing remote sensing data or severe cloud obstruction, multi-source image fusion or interpolation reconstruction algorithms can be introduced to improve spatial coverage. Simultaneously, in cases of significant regional scale differences, different spatial scale modeling methods can be used to construct overall urban heat island indicators and refined heat island indicators at the community or residential point level. In the long-term exposure modeling stage, exposure data can be weighted over multiple years, stratified seasonally, or weighted by extreme high-temperature events, based on the research period and the distribution characteristics of the research subjects, to enhance the model's ability to characterize the cumulative effects of chronic heat exposure. For individuals with incomplete residential information or those who have migrated, their exposure levels can be estimated through address repair, fuzzy matching, or weight redistribution methods, thereby ensuring the stability and continuity of long-term heat exposure measurement results.

[0028] The present invention discloses the following steps for matching and integrating baseline information, health and disease data, exposure data, and multi-omics data to establish a research and analysis database: The baseline information of the research subjects was checked for consistency and missing values. Records with logical conflicts, incompleteness or abnormal values ​​were marked, corrected or removed to ensure data quality. Further standardization of health outcome data involves converting disease diagnosis data, mortality information, and follow-up records from multiple sources or different coding systems into a unified standard disease coding format, and labeling the follow-up status of study subjects according to the time of event occurrence. Secondly, the environmental exposure data is spatially and temporally aligned. The urban thermal environment data is processed with uniform resolution according to the preset spatial units, and spatial matching is performed based on the residential address information of the research subjects to form environmental exposure indicators at the individual level. Based on individual identifiers of the study participants, baseline information, health and disease data, and matched environmental exposure data are integrated to generate an analytical database containing individual characteristics, outcome events, and long-term environmental exposure levels. Finally, the constructed research analytical database is versioned and structured for storage, providing a unified data foundation for subsequent model analysis and results output.

[0029] As an optional implementation, during the construction of the research and analysis database, the system can automatically map and restructure fields of data from different sources, converting unstructured text information into standardized variable formats suitable for modeling. To enhance data consistency, time information can be uniformly converted into standard timestamps, spatial information into a uniform latitude and longitude coordinate system, and a unified primary key mechanism ensures accurate correlation of multi-source data. For data records updated multiple times during the follow-up process, the system can use a version control mechanism to save data snapshots at each stage and automatically identify valid record intervals during the analysis phase. Regarding privacy protection, the system can anonymize or desensitize personal identification information before data is entered into the database, retaining only necessary variables for statistical analysis. A permission control module enables hierarchical management of database access, thereby ensuring the integrity of the data required for algorithm operation while improving the overall system's data security.

[0030] The present invention discloses the following steps for generating predictive feedback data under different intervention or mitigation scenarios by analyzing an integrated database: First, based on the integrated multi-omics data, urban heat island exposure levels, and baseline covariates, the system constructs a feature engineering module to uniformly map protein expression levels, metabolite intensity, heat island exposure values, and individual characteristics to the machine learning modeling space. Subsequently, in the artificial intelligence feature selection module, an Elastic Net regularized regression model is used to screen key molecular features significantly related to urban heat island stress from high-dimensional proteomics and metabolomics data, achieving dimensionality reduction and variable screening at the molecular level. The objective function is expressed as follows: ; in: For individuals At the molecular level (e.g., protein / metabolite strength). For the first A sample was exposed to the heat island or covariate. The value on, The regression coefficients to be estimated are: For regularization strength, This module controls the mixing ratio of Lasso regression and Ridge regression. It adaptively determines the optimal parameter combination within the AI ​​training framework using cross-validation, thereby improving generalization ability while controlling model complexity.

[0031] After obtaining the set of key molecular features, the system uses urban heat island exposure indicators, screened multi-omics molecular features, and individual baseline information as input variables to construct a health risk prediction model that integrates AI inference mechanisms. This embodiment employs a survival prediction model based on the Cox proportional hazards model, embedding machine learning algorithms for parameter optimization and model scheduling to construct a joint prediction framework between urban heat island, multi-omics, and health risk. Its risk function is expressed as follows: ; in :individual In time The instantaneous risk; Used as the benchmark risk function; Individual heat island exposure value; For the first A molecular signature associated with heat island exposure; Covariates (such as age, gender, BMI, etc.); : The corresponding regression coefficients. The artificial intelligence module continuously optimizes the model's stability and prediction accuracy through Bootstrap resampling, cross-validation, and automatic parameter tuning algorithms, and uses the consistency index (C-index) as the main performance evaluation indicator.

[0032] Based on this, the system uses an artificial intelligence scenario simulation module to simulate and analyze different intervention strategies, including but not limited to assuming a 5% or 10% reduction in urban heat island intensity, increasing urban green coverage, optimizing urban ventilation corridor structure, and implementing individualized cooling interventions. It constructs a multi-scenario prediction dataset and estimates the relative risk changes under different intervention conditions based on the risk function output by the model.

[0033] To quantitatively assess the actual health effects of interventions, this invention introduces an absolute risk reduction (ARR) index to measure the magnitude of change in health risk under different scenario assumptions. Its calculation formula is as follows: ; in, The predicted risk of morbidity or mortality under the reference scenario (no intervention); This system predicts risks under intervention scenarios. It uses an AI decision engine to rank and compare the ARR (Advanced Risk Ratio) of different strategies, automatically identifying the intervention plan with the greatest health benefit.

[0034] As an optional implementation, during the model construction process, the system can dynamically select different types of machine learning algorithms based on data scale, variable dimensionality, and complexity, forming an integrated modeling framework with the statistical model. For example, in the feature selection stage, random forests, gradient boosting trees, or deep learning networks can be introduced to rank the importance of variables, assisting traditional regression models in feature reduction; in the risk prediction stage, neural network survival models or XGBoost survival models can enhance the ability to characterize nonlinear relationships. During parameter optimization, the system can dynamically adjust model parameters through automated parameter tuning algorithms and continuously monitor model stability during training. When new data is continuously added, the system can automatically trigger incremental training or full retraining mechanisms to ensure that the prediction model always maintains its adaptability to changes in data distribution. In the scenario simulation stage, the system can support batch generation of virtual exposure strategy combinations and automatically call the model for risk reassessment, thereby quickly comparing the long-term impact of different intervention paths on health outcomes.

[0035] The present invention discloses the following steps for outputting individualized or group-based intervention effect evaluation results based on model feedback data: The system further invokes the artificial intelligence decision support module to transform the model prediction results into structured feedback information; On the one hand, it generates personalized risk assessment reports for individuals, including the absolute risk of disease within a specified time window, the change in risk before and after intervention, and the stress response of key molecular pathways; On the other hand, policy evaluation reports are generated for the population, quantifying the number of cases or deaths that different urban heat island mitigation measures can avoid at the population level, and providing decision-makers with data-driven intervention recommendations and prioritization results.

[0036] As an optional implementation, the system of this invention can automatically generate different levels of result presentation methods according to user type. For example, it can output concise health risk warnings and intervention suggestions for individual users, complete model parameters, confidence intervals, and visualization charts for research institutions, and quantifiable health benefit assessment indicators and intervention priority ranking results for government management departments. In terms of data display, the system can dynamically display risk change curves, spatial distribution maps, and scenario comparison results through a graphical interface, thereby enabling an intuitive interpretation of complex model results. For key prediction results, the system can also provide an explanation of the uncertainty interval to assist decision-makers in assessing the reliability of risk predictions and avoiding misleading the decision-making process with a single result.

[0037] On the other hand, embodiments of the present invention also provide a system for implementing the above-mentioned method for urban heat island stress identification, health risk prediction, and intervention effect evaluation based on artificial intelligence multi-omics fusion. This system is a health risk prediction and intervention evaluation system based on urban heat island stress identification, referencing... Figure 2 ,include: The data collection module is used to collect baseline information of research subjects and continuously acquire health and disease information, urban heat island exposure data, and multi-omics data such as proteomics and metabolomics during the follow-up period through cohort study design, follow-up function module or short-term intervention trial. It also supports automatic access to environmental information from satellite remote sensing, meteorological station network and urban spatial database.

[0038] As an optional implementation, the data collection module of this invention can collect data through various interfaces, including hospital business system interfaces, public health platform interfaces, remote sensing data interfaces, and questionnaire system interfaces, and supports automatic identification and parsing of data formats from different data sources. For health and disease information, it can be automatically updated through inpatient records, outpatient systems, death registration systems, etc.; for multi-omics data, structured result files can be directly imported from experimental testing platforms or third-party testing systems. For urban heat island exposure data, temperature and surface parameter information can be periodically obtained through satellite remote sensing platforms and updated in real time in conjunction with meteorological monitoring station data. The system supports automatic data collection according to predetermined time windows and also supports manual triggering of collection mechanisms to meet the data update needs of different scenarios.

[0039] The data processing module is used to perform automated preprocessing operations on the collected multi-source heterogeneous data, including anomaly identification, missing value handling, batch effect correction, unit unification, time format standardization, and spatial coordinate transformation. Based on intelligent data fusion algorithms, it performs spatiotemporal matching and multimodal alignment of environmental exposure data, molecular data, and individual health data to construct a research and analysis database with a unified structure.

[0040] As an optional implementation, the data processing module of this invention can operate collaboratively with a rule engine and an algorithm engine to perform intelligent verification on data from different sources, including detecting time conflicts, duplicate records, abnormal fluctuation values, and missing fields, and automatically marking or removing data that does not meet quality standards. Regarding identity association, a unified identifier key is used to link and integrate information about the same research subject from different data sources; regarding spatiotemporal alignment, a unified timestamp and spatial coordinate matching mechanism ensures the consistency of environmental exposure data and individual information in both time and space. The system can also automatically calculate derived variables based on user settings, such as long-term exposure mean, cumulative exposure, and age-standardized variables, ensuring that the data directly meets modeling requirements. After the above preprocessing and fusion steps, the system can perform joint modeling of urban heat island exposure information and proteomics characteristics at a unified spatial and temporal scale, and identify key protein signals significantly related to urban heat island stress. For example, such as... Figure 3 As shown, the multi-omics fusion analysis process constructed based on the embodiments of the present invention can screen out protein expression features that are significantly related to urban heat island exposure levels from massive proteomics data. This result demonstrates the automatic identification capability of the present invention for molecular response signals under complex environmental exposure factors, providing key biomarker inputs for subsequent health risk prediction models.

[0041] The artificial intelligence analysis module is used to embed machine learning and deep learning algorithms to perform feature engineering, variable screening, model training and parameter optimization operations on the exposure-molecular-health outcome data in the research analysis database, so as to realize the identification of key molecular features, construction of health risk prediction models and dynamic updating of models.

[0042] As an optional implementation, the artificial intelligence analysis module of this invention can automatically select appropriate modeling strategies based on the data scale and complexity, including but not limited to regularized regression, tree models, neural networks, or ensemble learning algorithms, and supports parallel model training and ensemble model fusion. During the model parameter optimization stage, automatic parameter tuning algorithms can be embedded, and cross-validation and model stability monitoring mechanisms are introduced to automatically identify overfitting risks and optimize the model structure. For intermediate results and final parameters generated during model operation, the system supports automatic storage and version tracking, making the model interpretable and reproducible. When new data enters the system, the artificial intelligence module can automatically initiate incremental learning or model retraining functions, enabling the algorithm to dynamically evolve as data is updated.

[0043] The scenario simulation and feedback module is used to build a virtual intervention environment based on artificial intelligence prediction models, simulate and extrapolate the changes in urban heat island intensity under different mitigation schemes or intervention strategies, and output multi-dimensional feedback data such as changes in individual disease risk, key molecular stress changes, and trends in health outcome improvement under the corresponding conditions.

[0044] As an optional implementation, the scenario simulation module of this invention supports the construction of a multi-dimensional intervention space, allowing users to input different intervention conditions in a parameterized manner, such as the urban surface cooling rate, the percentage increase in green coverage, the level of improvement in ventilation conditions, or the intensity of individual cooling intervention. The system can automatically combine multiple types of parameters to generate a multi-scheme simulation cluster and call the artificial intelligence analysis module to perform batch risk simulations on each simulation scenario. The feedback results are output in the form of structured data, including the risk change trend under different intervention scenarios, the response of key molecules, and the scale of population benefit, providing a computational basis for the subsequent decision-making module.

[0045] The decision support module is used to comprehensively analyze and quantitatively evaluate the model output results, transforming the predicted risk level, absolute risk reduction indicators, and population distribution structure information into individual health management suggestions and group intervention priority plans, and outputting intelligent decision support results.

[0046] As an optional implementation, the decision support module of this invention can automatically adapt the output mode according to different user identities. For example, it can generate health risk warnings and intervention suggestion summaries for individual users, output complete model parameters and uncertainty ranges for researchers, and output quantifiable health benefit assessment results and intervention priority ranking schemes for management departments. The system can also perform self-learning based on historical intervention effect data, and provide feedback correction on the actual effects of implemented intervention measures, so that the decision support results have adaptive update capabilities, improving the reliability and scientific nature of decision-making.

[0047] This invention enables measurable and controllable management of the heat exposure-molecular-health chain through high-precision quantification of individual heat exposure, multi-omics-driven stress identification, and causal-oriented risk prediction, thereby demonstrating clear clinical and public health benefits in terms of intervention effects. First, this system transforms intervention effects from traditional subjective feelings and crude health indicators into a multi-level quantitative chain of "exposure reduction – biomarker response – improved health outcomes": it can directly measure the magnitude of individual exposure reduction caused by the intervention, differential changes in key inflammatory / metabolic biomarkers, and model-based reductions in short- and medium-term disease risk scores. Based on these quantifiable indicators, this invention can not only demonstrate the effectiveness of single interventions (such as community greening, home cooling optimization, or nutritional intervention) at the molecular and physiological levels, but also assess the synergistic effects of combined interventions, thus providing evidence to support the development of the most cost-effective interventions. Furthermore, this invention has direct application value and observable socio-economic effects in reducing the burden on the health system and society. It not only technically realizes the connection from individual molecular evidence to group health and policy decisions, but also improves the accuracy and verifiability of interventions. It also provides empirical tools and methods for reducing the medical burden, optimizing the allocation of public resources, and enhancing urban climate adaptability.

[0048] Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications and improvements without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the present invention shall be defined by the claims.

Claims

1. A health risk prediction and intervention assessment system based on urban heat island stress identification, characterized in that, include: The data collection module is used to collect baseline information of research subjects, continuously acquire health and disease information, urban heat island exposure data, and proteomics and metabolomics data during the follow-up period, and also supports automatic access to multi-source environmental data from satellite remote sensing, meteorological station networks and urban spatial information platforms. The data processing module is used to perform automated preprocessing and data standardization operations on the collected multi-source heterogeneous data, and based on intelligent data fusion algorithms, it performs spatiotemporal matching and multimodal alignment of environmental exposure information, molecular level information and individual health information to construct a research and analysis database with a unified structure. The artificial intelligence analysis module is used to embed machine learning and deep learning algorithm frameworks to automatically perform feature engineering, variable selection and model training on exposure-molecule-health information in the database. It completes the screening of key molecular features of multi-omics, the construction of health risk prediction models and adaptive optimization of parameters, and continuously updates model weights based on algorithm tuning mechanism to achieve dynamic learning and iterative optimization of model performance. The scenario simulation and feedback module is used to build a virtual intervention environment based on the artificial intelligence prediction model, simulate and extrapolate the changes in urban heat island intensity under different mitigation scenarios, and output the changing trends of individual risk levels, the magnitude of changes in molecular stress response and the changes in the probability of disease occurrence under various assumptions, forming multi-dimensional prediction and feedback data. The decision support module is used to intelligently analyze and sort the above-mentioned forecast feedback data, and combine it with changes in risk value, absolute benefit indicators and population distribution structure to generate individualized health management suggestions and group intervention priority plans. It also outputs policy evaluation reports, risk classification lists or intervention benefit interpretation documents through a visual interface.

2. The health risk prediction and intervention assessment system based on urban heat island stress identification as described in claim 1, characterized in that, The data collection module integrates satellite remote sensing, meteorological monitoring and urban morphology databases based on baseline information and spatial location data to construct an urban heat island exposure model covering the study area and generate corresponding individualized heat exposure indicators for each study object.

3. The health risk prediction and intervention assessment system based on urban heat island stress identification as described in claim 2, characterized in that, The data collection module specifically includes: First, acquire multi-source remote sensing observation data, including medium- and high-resolution land surface temperature, normalized vegetation index, land surface albedo, land use and land cover data, and perform quality control and spatiotemporal registration on them. Subsequently, surface characteristic indicators were used to identify urban areas, and a stable urban boundary mask was constructed by combining administrative boundaries and light and night remote sensing data. After obtaining the urban area, the background temperature was calculated using the urban-suburban comparison method; Finally, the land surface temperature of each grid point in the urban area is compared with its corresponding background temperature. The heat island intensity value is obtained by calculating grid by grid, and an individualized urban heat island exposure index is generated for each research subject based on its exposure location and activity time.

4. The health risk prediction and intervention assessment system based on urban heat island stress identification as described in claim 3, characterized in that, The formula for calculating the intensity value of the heat island is as follows: ； in, For urban areas in spatial location and time Surface or temperature value; The background temperature is the same as that of a reference station in the suburbs or rural areas during the same period.

5. The health risk prediction and intervention assessment system based on urban heat island stress identification as described in claim 1, characterized in that, The data processing module performs intelligent verification on data from different sources, including detecting time conflicts, duplicate records, abnormal fluctuation values, and missing fields, and automatically marks or removes data that does not meet quality standards. In terms of identity association, it integrates and associates information of the same research subject from different data sources through a unified identification key. In terms of spatiotemporal alignment, it ensures consistency between environmental exposure data and individual information in terms of time and space through a unified timestamp and spatial coordinate matching mechanism.

6. The health risk prediction and intervention assessment system based on urban heat island stress identification as described in claim 5, characterized in that, The data processing module specifically includes: The baseline information of the research subjects was checked for consistency and missing values. Records with logical conflicts, incompleteness or abnormal values ​​were marked, corrected or removed. The health outcome data were standardized by converting disease diagnosis data, death information and follow-up records from multiple sources or different coding systems into a standard disease coding format, and the follow-up status of the study subjects was marked according to the time of the event. The environmental exposure data is spatially and temporally aligned, the urban thermal environment data is processed with uniform resolution according to the preset spatial units, and spatial matching is performed based on the residential address information of the research subjects to form environmental exposure indicators at the individual level. Based on the individual identifiers of the study subjects, baseline information, health and disease data, and matched environmental exposure data are integrated to generate an analytical database that includes individual characteristics, outcome events, and long-term environmental exposure levels. Finally, the completed research and analysis database is versioned and stored in a structured manner.

7. The health risk prediction and intervention assessment system based on urban heat island stress identification as described in claim 1, characterized in that, The artificial intelligence analysis module performs standardization, missing value imputation, outlier identification, and variable recoding operations on proteomics and metabolomics data, converting the original multidimensional molecular signals into a feature expression space that can be learned by the model. Based on the integrated multi-omics data, urban heat island exposure level, and baseline covariates, a feature engineering module is constructed to uniformly map protein expression levels, metabolite intensity, heat island exposure value, and individual characteristics to the machine learning modeling space.

8. The health risk prediction and intervention assessment system based on urban heat island stress identification as described in claim 1, characterized in that, The artificial intelligence analysis module employs an Elastic Net regularized regression model to screen key molecular features significantly associated with urban heat island stress from high-dimensional proteomics and metabolomics data, achieving dimensionality reduction and variable screening at the molecular level. Its objective function is expressed as: ； in: For individuals At the molecular level, For the first A sample was exposed to the heat island or covariate. The value on, The regression coefficients to be estimated are: For regularization strength, Control the mixing ratio of Lasso regression and Ridge regression.

9. The health risk prediction and intervention assessment system based on urban heat island stress identification as described in claim 8, characterized in that, After obtaining the key molecular feature set, the artificial intelligence analysis module uses urban heat island exposure indicators, screened multi-omics molecular features, and individual baseline information as input variables. A health outcome prediction network is constructed with a survival analysis model at its core. Machine learning mechanisms are used to optimize variable combinations and update model parameters, achieving joint modeling of urban heat island exposure, multi-omics molecular features, and health risk. Its risk function is expressed as: ； in :individual In time The instantaneous risk; Used as the benchmark risk function; Individual heat island exposure value; For the first A molecular signature associated with heat island exposure; For covariates; The corresponding regression coefficients obtained through maximum likelihood iterative optimization are trained by the artificial intelligence module.

10. The health risk prediction and intervention assessment system based on urban heat island stress identification as described in claim 1, characterized in that, The artificial intelligence analysis module introduces an absolute risk reduction index, the calculation formula of which is: ； in, The predicted risk of morbidity or mortality under a reference scenario; The decision support module predicts the risks under the intervention scenario; it ranks and optimizes each intervention plan based on the absolute risk reduction index and the overall risk change trend, and further converts them into the number of avoidable cases, the number of avoidable deaths, or the reduction in molecular damage.

Images