An ada boost-based urban temperature inference and heat exposure risk assessment method

Abstract

The present application relates to the technical field of air temperature risk assessment, and particularly relates to a city air temperature inference and heat exposure risk assessment method based on AdaBoost. The method comprises the following steps: obtaining original city form data and city meteorological data of the city; performing data preprocessing on the original city form data and the city meteorological data of the city to generate processed original city form data and city meteorological data; integrating the processed original city form data and the city meteorological data into a model training set and a model test set; designing an artificial neural network architecture by using an AdaBoost algorithm, and performing model training on the model training set by using the artificial neural network architecture to generate a city air temperature inference pre-model. The present application improves the accuracy and reliability of city air temperature inference and heat exposure risk assessment by fusing multi-source data, applying advanced machine learning algorithms, and strengthening sensitivity analysis and Monte Carlo simulation.

Description

Technical Field

[0001] This invention relates to the field of temperature risk assessment technology, and in particular to a method for urban temperature inference and heat exposure risk assessment based on AdaBoost. Background Technology

[0002] Since the mid-20th century, the urban heat island effect has become a key focus of climatological and environmental research. Scholars have found that temperatures in urban areas are generally higher than in surrounding rural areas, leading to increased heat exposure risk, especially during periods of high temperatures. Early studies relied primarily on surface meteorological observation data for localized temperature estimations, but due to the complexity and variability of cities, traditional methods struggled to provide a comprehensive and accurate assessment. With the development of remote sensing technology and Geographic Information Systems (GIS), the spatial distribution and dynamic changes of urban temperatures have been monitored with greater precision. Remote sensing images, using the thermal infrared band, can acquire surface temperature data. Combined with GIS technology, this allows for temperature inference and spatial distribution analysis of different urban areas, helping to identify high-risk areas for heat exposure. In the 21st century, the application of climate models and machine learning methods has significantly improved urban temperature inference and heat exposure risk assessment methods. However, current technologies often rely on a single data source, such as urban meteorological data or urban morphology data, neglecting the coupling effect between the two. Furthermore, they typically depend on simple statistical regression models, failing to capture complex nonlinear relationships, resulting in low prediction accuracy and low reliability of risk assessments. Summary of the Invention

[0003] Therefore, it is necessary to provide a method for urban temperature inference and heat exposure risk assessment based on AdaBoost to solve at least one of the above-mentioned technical problems.

[0004] To achieve the above objectives, a method for urban temperature inference and heat exposure risk assessment based on AdaBoost is proposed, the method comprising the following steps:

[0005] Step S1: Obtain the original urban morphology data and urban meteorological data of the city; perform data preprocessing on the original urban morphology data and urban meteorological data of the city to generate processed original urban morphology data and urban meteorological data; integrate the processed original urban morphology data and urban meteorological data into the model training set and the model test set.

[0006] Step S2: Design an artificial neural network architecture using the AdaBoost algorithm, and use the artificial neural network architecture to train the model on the training set to generate a pre-model for inferring urban temperature; validate the pre-model for inferring urban temperature based on the test set to generate a model for inferring urban temperature.

[0007] Step S3: Perform global sensitivity analysis and local correlation analysis on the parameters in the urban temperature inference model to generate global sensitivity data and local correlation data; confirm the heat exposure risk threshold based on the global sensitivity data and local correlation data to obtain the heat exposure risk threshold;

[0008] Step S4: Import the processed original urban morphology data and urban meteorological data into the urban temperature inference model for Monte Carlo simulation to generate temperature simulation samples; identify risk hotspots in the temperature simulation samples based on the heat exposure risk threshold to generate urban heat exposure risk area data.

[0009] This invention ensures data quality and consistency by acquiring and preprocessing urban morphology and meteorological data, providing a reliable foundation for subsequent model training and prediction. The artificial neural network architecture designed using the AdaBoost algorithm effectively improves the model's prediction accuracy and generalization ability, especially performing exceptionally well when handling complex nonlinear relationships. The prediction model is validated using a test set, ensuring its reliability and stability and avoiding overfitting or underfitting. Global sensitivity analysis and local correlation analysis help identify key parameters in the model, further optimizing model performance and providing a scientific basis for confirming heat exposure risk thresholds. Based on global sensitivity data and local correlation data, heat exposure risk thresholds can be determined more accurately, providing a clear reference standard for urban thermal environment management. This threshold can be used to identify high-risk areas, helping decision-makers formulate targeted response measures. A large number of temperature simulation samples are generated through Monte Carlo simulation, comprehensively covering various temperature change scenarios and enhancing the robustness of the prediction results. Combined with heat exposure risk thresholds, heat exposure risk areas in cities can be effectively identified, providing important references for urban planning, emergency management, and public health. This method is not only applicable to current urban thermal environment analysis, but can also be applied to heat exposure risk assessment under future climate change scenarios by updating data and models. Its modular design (such as data preprocessing, model training, and simulation analysis) makes it highly flexible and scalable, allowing for adjustments and optimizations to meet the needs of different cities. Therefore, this invention improves the accuracy and reliability of urban temperature inference and heat exposure risk assessment by fusing multi-source data, applying advanced machine learning algorithms, and enhancing sensitivity analysis and Monte Carlo simulation.

[0010] Preferably, obtaining the original urban morphology data and urban meteorological data of the city includes:

[0011] Obtain the geographical location of the city; use 3D modeling technology to extract the building height, building coverage, exterior wall area, proportion of hard paving, and ground reflectivity of the geographical location of the city to obtain urban morphological characteristic data;

[0012] The sky view factor is calculated by using a fisheye camera to measure the geographical location of the city. The sky view factor is then integrated with the city morphology data to obtain the original city morphology data.

[0013] Reference weather stations were set up based on the original urban morphology data of the city.

[0014] Based on the reference weather station, a mobile weather station is installed at the pedestrian height point, and the mobile weather station is used to collect air temperature, relative humidity, wind speed and solar radiation; thus, urban meteorological data is obtained.

[0015] This invention ensures targeted data collection and complete regional coverage by clearly defining the geographical location of the city. Utilizing 3D modeling technology to extract building height, building coverage, exterior wall area, proportion of hard paving, and ground reflectivity, it accurately quantifies the city's three-dimensional morphological characteristics, providing detailed spatial data support for subsequent thermal environment analysis. Calculating the sky visibility factor using a fisheye camera accurately reflects the city's sky visibility (SVF), a crucial factor influencing urban thermal environments (such as the heat island effect). Combining building morphology data with the sky visibility factor forms comprehensive raw urban morphology data, enabling a more accurate description of the impact of urban spatial structure on the thermal environment. Setting up reference meteorological stations based on urban morphology data ensures the scientific validity and representativeness of meteorological data collection, avoiding data bias. Installing mobile meteorological stations at pedestrian height allows for the collection of microclimate data directly related to human thermal perception (such as air temperature, relative humidity, wind speed, and solar radiation), improving the data's practicality and relevance. Simultaneously collecting multi-dimensional meteorological data such as temperature, humidity, wind speed, and solar radiation provides comprehensive input parameters for urban thermal environment analysis. By utilizing mobile weather stations and fisheye camera technology, high-resolution urban morphology and meteorological data can be acquired at a local scale, overcoming the limitations of traditional weather station data in terms of spatial resolution. This high-resolution data enables more refined urban thermal environment analysis, such as identifying local heat island effects or microclimate differences. The acquired high-precision urban morphology and meteorological data provide reliable input for subsequent urban temperature inference models, improving their prediction accuracy. By combining urban morphology and meteorological data, urban heat exposure risks can be assessed more accurately, providing a scientific basis for urban planning and management.

[0016] Preferably, the calculation of the sky view factor for the geographical location of the city using a fisheye camera includes:

[0017] The city's sky image is generated by taking pictures of the geographical area of ​​the city using a fisheye camera.

[0018] Orientation analysis is performed on urban sky images to generate urban sky orientation data; based on the urban sky orientation data, urban sky images are partitioned to generate urban sky partition image sets;

[0019] Sky visibility is calculated for the urban sky partition image set to obtain the partition sky visibility; the urban sky partition image is then used to perform skyline contour detection to generate partition sky skyline contour data.

[0020] The sky view factor is calculated for the urban sky zone image set based on the zoned sky visibility and zoned sky skyline contour data. The formula for calculating the sky view factor is as follows:

[0021]

[0022] in, For the sky view factor, This refers to the visible area of ​​the sky in a photograph of the sky hemisphere. This represents the area of ​​the entire sky hemisphere image.

[0023] This invention provides scientific and objective skyscape scores for different regions based on three core factors: sky visibility, obstruction, and skyline complexity. This contributes to the optimization of urban airspace in fields such as urban planning and environmental design. By zoning the urban sky, it enables detailed analysis of skyscapes from different geographical locations and directions, identifying areas with good skyscapes and those affected by tall buildings or pollution. This method helps urban designers and environmental planners better assess sky openness and its impact on the urban environment, thereby optimizing building layout, green space planning, and improving residents' quality of life. Skyline detection reflects the city's building density and structural characteristics. Combined with other data, it provides richer information for urban visualization modeling, aiding in the analysis and improvement of the urban landscape. The data generated through fisheye camera photography and sky visibility factor calculation not only serves as basic theoretical support but also provides a basis for subsequent urban development, landscape design, and green space planning.

[0024] Preferably, step S2 includes the following steps:

[0025] Step S21: Based on the random forest permutation importance method, key variables are screened for the processed original urban morphology data and urban meteorological data to generate key variables for urban morphology and key variables for urban temperature;

[0026] Step S22: Introduce model interaction terms; design an artificial neural network architecture based on the AdaBoost algorithm, wherein the artificial neural network architecture includes a base learner and an ensemble strategy;

[0027] Step S23: Determine the feature vectors of the input layer neurons of the artificial neural network architecture based on the key variables of urban morphology, key variables of urban temperature, and model interaction terms;

[0028] Step S24: Use the feature vectors of the input layer neurons to perform a grid search on the artificial neural network architecture to confirm the optimal parameters; train the model on the training set based on the optimal parameters of the artificial neural network architecture to generate a pre-model for inferring urban temperature; validate the pre-model for inferring urban temperature based on the test set to generate a model for inferring urban temperature.

[0029] This invention, through the random forest importance ranking method in step S21, can screen out the most critical variables related to urban temperature and urban morphology. This avoids data redundancy, improves model efficiency and accuracy, and thus helps optimize the urban temperature prediction model. By introducing model interaction terms and combining them with an artificial neural network architecture designed using the AdaBoost algorithm (step S22), the model's ability to model complex nonlinear relationships is enhanced, improving the accuracy of temperature prediction. This ensemble learning method is particularly effective when dealing with different types of data, improving overall prediction stability. By determining the neuron feature vectors and using grid search to find optimal parameters in step S24, the artificial neural network can fully utilize urban morphology and meteorological data, improving the accuracy and efficiency of model training. In this way, the model can adapt to different climatic conditions and urban environments, achieving more accurate temperature predictions. By training the model on the training set and validating it on the test set, the reliability and generalization ability of the prediction model are ensured. This makes the urban temperature inference model applicable not only to current data but also capable of handling future changes, improving the model's long-term application value. The generated temperature inference model can provide a scientific basis for urban planning and environmental regulation. For example, in the study of the urban heat island effect, suggestions for optimizing urban morphology can be based on model-predicted temperature changes, thereby achieving sustainable urban development and environmental protection.

[0030] Preferably, an artificial neural network architecture is designed based on the AdaBoost algorithm, wherein the artificial neural network architecture includes a base learner and an ensemble strategy including:

[0031] Configure the artificial neural network; configure the hidden layers of the model based on the ReLU activation function, where the formula for the ReLU activation function is shown below:

[0032]

[0033] in As the dependent variable, As the independent variable, The unstandardized coefficients of the independent variable. For the number of parameters, This is the error term; This represents the index range of the independent variables in the regression equation.

[0034] The model output layer is configured based on a linear function, and the hidden layer and the model output layer are integrated as base learners.

[0035] An ensemble strategy is generated by adaptively weighting and cascading base learners using the AdaBoost algorithm.

[0036] This invention utilizes the ReLU activation function (the formula in the steps provides the basic form of the OLS regression model) to provide nonlinear mapping capabilities to the hidden layers. The ReLU function, by maintaining linearity in the positive interval and outputting zero in the negative interval, enables the neural network to better capture complex nonlinear relationships in the data. For complex problems such as temperature prediction, this nonlinear characteristic helps improve the model's ability to learn complex patterns. In the model's output layer configuration, the linear function setting allows the network to produce linear regression results that conform to the actual data. With this configuration, the artificial neural network can handle complex nonlinear features in the hidden layers and convert these features into accurate prediction results through the linear output layer. This structure combines nonlinear fitting capabilities with linear output capabilities, achieving a balance across various tasks. Weighted cascade ensemble integration of the base learners using the AdaBoost algorithm further improves the model's prediction accuracy. AdaBoost enhances the performance of weak learners through an iterative weighting process. Weak learners pay more attention to data samples that were not correctly predicted in the previous round in each iteration, enabling the ensemble model to better handle complex and dynamic data patterns. In urban temperature prediction, this effectively improves the model's robustness to changing environments. Because of AdaBoost's adaptive weighting mechanism, the weights of each base learner can be dynamically adjusted based on their error magnitude. This makes the ensemble model more inclined to focus on samples with larger prediction errors, avoiding overfitting of individual base learners on the training data, thus improving the model's generalization ability. This is particularly important for temperature prediction models, as climate data itself has high noise and volatility. By combining base learners with the AdaBoost algorithm, the model can automatically optimize during each training iteration. Unlike traditional single neural network training methods, AdaBoost can progressively optimize the learner's performance over multiple iterations, achieving higher prediction accuracy when dealing with complex data. This allows the method to flexibly adapt to different types of datasets and provide relatively stable performance. The ensemble strategy not only improves the accuracy of individual base learners but also enhances the overall model's adaptability to different data distributions and climate change. Through weighting and ensemble, the model can automatically adjust according to the actual performance of the data, improving its effectiveness in practical applications. For example, urban temperature changes are often influenced by multiple factors; the ensemble strategy can effectively consider these diverse influencing factors, providing more accurate predictions.

[0037] Preferably, step S23 includes the following steps:

[0038] Step S231: Perform feature vector transformation on the key variables of urban morphology and urban temperature to generate urban morphology feature vector and urban temperature feature vector; perform correlation analysis on the urban morphology feature vector and urban temperature feature vector to generate morphology-temperature correlation data.

[0039] Step S232: Calculate the feature weights of the morphology-temperature correlation data to obtain feature weight data; optimize the feature combination of the morphology-temperature correlation data based on the feature weight data according to the model interaction terms to generate combined feature data;

[0040] Step S233: Perform dimensional transformation based on the combined feature data to obtain transformed dimensional data; initialize neurons using the transformed dimensional data to obtain the feature vector of the input layer neurons.

[0041] This invention reveals the potential relationship between urban morphology and temperature by performing correlation analysis on urban morphology feature vectors and urban temperature feature vectors. This correlation analysis helps understand how different urban morphologies (such as building density and exterior wall area) affect local temperature, providing theoretical support for subsequent model optimization. Feature weight calculation and feature combination optimization ensure that the final model input focuses on the features most influential on the prediction results. This feature weighting optimization helps reduce the interference of redundant features on model performance while increasing the weight of key features in the model. Through feature combination optimization based on model interaction terms, more complex nonlinear feature combinations can be discovered, improving the model's prediction accuracy. By performing dimensionality transformation on the combined feature data and initializing neurons, more representative and discriminative input features can be provided for artificial neural networks. These dimensionally transformed and optimized feature vectors can better represent the complex relationship between urban morphology and temperature, thereby improving the learning ability of the neural network. In the case of multiple features, excessively high dimensionality can lead to overfitting or increased computational complexity. By optimizing dimensionality transformation and feature combination, redundant information can be effectively reduced, avoiding the curse of dimensionality and thus improving the computational efficiency and accuracy of the model. This is particularly important for large-scale urban temperature prediction models, as it significantly enhances computational efficiency. Initializing the feature vectors of input layer neurons ensures that the neural network can begin learning from an optimal starting point, avoiding slow convergence or getting stuck in local optima during the early stages of training. The optimized feature vectors enable the neural network to more quickly capture the complex patterns between urban morphology and temperature.

[0042] Preferably, step S3 includes the following steps:

[0043] Step S31: Calculate the full-order sensitivity index of the parameters in the urban temperature inference model based on the Sobol method to generate global sensitivity data;

[0044] Step S32: Perform local correlation analysis on the parameters in the urban temperature inference model to generate local correlation data. The local correlation analysis includes urban morphology factor-temperature nonlinear relationship analysis and reference meteorological parameter-temperature trend analysis.

[0045] Step S33: Confirm the high-risk threshold for heat exposure based on global sensitivity data and local correlation data to obtain the heat exposure risk threshold.

[0046] This invention utilizes the Sobol method to calculate the full-order sensitivity index, enabling the identification of the most critical parameters in urban temperature inference models. Sensitivity analysis of model parameters reveals which factors have the greatest impact on temperature changes and which factors have a smaller impact on model predictions. This simplifies the model, reduces unnecessary complexity, and focuses on the most important influencing factors. Local correlation analysis, through the analysis of the nonlinear relationship between urban morphology factors and temperature, and the analysis of reference meteorological parameters and temperature trends, reveals the complex relationships between different environmental factors and temperature. In particular, the nonlinear relationship between landscape factors (such as urban green space and building density) and temperature helps identify the micro-characteristics of temperature changes. Trend analysis of wind speed and temperature helps understand the moderating role of wind speed in different urban environments, especially under high-temperature conditions, where wind speed becomes a key factor influencing temperature. Based on global sensitivity data and local correlation data, a more accurate high-risk threshold for heat exposure can be identified. This risk threshold not only considers the sensitivity of each factor but also adjusts the risk assessment based on specific local correlation data, making heat exposure warnings more accurate and practical. In this way, extreme high-temperature weather can be effectively prevented and addressed, reducing the health risks posed by the urban heat island effect. Determining heat exposure risk thresholds can provide decision support for urban planners and policymakers. For example, in high-risk areas, priority can be given to planning more public cooling facilities or improving ventilation design, thereby effectively reducing the health risks posed by urban heat. Accurately identifying high-risk areas allows for more precise climate adaptation strategies for cities. For high-risk areas, targeted heatstroke prevention and cooling measures can be implemented, such as improving urban green spaces, optimizing building design, and enhancing the heat radiation control capabilities of buildings, to reduce the impact of excessively high temperatures on residents' health.

[0047] Preferably, step S32 includes the following steps:

[0048] Step S321: Extract the sky view factor, exterior wall area, hard paving ratio, and ground reflectivity from the urban temperature inference model, and discretize the sky view factor, exterior wall area, hard paving ratio, and ground reflectivity to generate the change values ​​of the sky view factor, exterior wall area, hard paving ratio, and ground reflectivity.

[0049] Step S322: Use the urban temperature inference model to predict temperature changes for each selected sky view factor, exterior wall area, ground hard paving ratio, and ground reflectivity, and generate a comprehensive impact factor-temperature change prediction map; perform slope change trend analysis on the comprehensive impact factor-temperature change prediction map to generate urban morphology factor-temperature nonlinear relationship data.

[0050] Step S323: Extract the reference point wind speed, reference point temperature, reference point solar radiation, and reference point humidity from the urban temperature inference model, and perform value change analysis to generate wind speed change values, reference point temperature change values, reference point solar radiation change values, and reference point humidity change values; use the urban temperature inference model to predict temperature changes for each selected wind speed change value, reference point temperature change value, reference point solar radiation change value, and reference point humidity change value, and generate a reference meteorological parameter-temperature change prediction map;

[0051] Step S324: Perform a downward trend analysis on the reference meteorological parameter-temperature change prediction map to generate negative correlation trend data of reference meteorological parameter-temperature change; integrate the urban morphology factor-temperature nonlinear relationship data and the reference meteorological parameter-temperature change negative correlation trend data into local correlation data.

[0052] This invention extracts and discretizes sky visibility factors, exterior wall area, hard paving ratio, and ground reflectivity to generate change values ​​for these factors, providing multi-dimensional input for subsequent temperature change prediction. This process comprehensively considers various factors of the built environment, thereby fully assessing their impact on urban temperature and ensuring more accurate temperature change prediction. The urban temperature inference model is used to predict temperature changes based on changing sky visibility factors, exterior wall area, hard paving ratio, and ground reflectivity. Through slope trend analysis, nonlinear relationship data between urban morphology factors and temperature is generated. This step reveals the complex influence of urban morphology on temperature, providing a theoretical basis for further urban design and climate adaptability research. Changes in meteorological parameters such as reference point wind speed, reference point temperature, solar radiation, and humidity in the urban temperature inference model are extracted and analyzed, generating change values ​​for each meteorological parameter. This data provides a rich source of information for further temperature change prediction, improving the accuracy and reliability of the prediction. By analyzing the negative correlation trend between reference meteorological parameters and temperature changes, negative correlation trend data between reference meteorological parameters and temperature changes is generated. This analysis clarifies which meteorological factors lead to temperature decreases, providing guidance for future urban temperature management and regulation. By integrating data on the nonlinear relationship between urban morphology factors and temperature, as well as data on the negative correlation trend between reference meteorological parameters and temperature changes, local correlation data was obtained. This integration provides multi-faceted and multi-layered support for temperature prediction models, enhancing the models' comprehensiveness and practical application value.

[0053] Preferably, step S33 includes the following steps:

[0054] Step S331: Identify sensitive factors in the global sensitivity data to obtain sensitive factor data; perform parameter response analysis on the global sensitivity data based on the sensitive factor data to obtain response feature data;

[0055] Step S332: Perform sensitivity quantification calculation on the response feature data to obtain quantitative index data; perform spatial correlation analysis on the local correlation data to obtain correlation feature data; perform multivariate coupling analysis on the response feature data and correlation feature data to obtain coupling effect data;

[0056] Step S333: Calculate the influence weights based on the coupling effect data to obtain weight data; use the weight data to identify the critical values ​​of the coupling effect data to obtain critical value data, and extract the highest value of the critical value data as the heat exposure risk threshold.

[0057] This invention identifies parameters that significantly influence urban temperature inference models through sensitivity factor identification. This identification helps understand which factors play a dominant role in heat exposure risk assessment, providing key variables for subsequent analysis. Parameter response analysis further reveals the response characteristics of different sensitivity factors to temperature changes, facilitating the accurate identification of key factors affecting urban temperature. Sensitivity quantification calculations quantify the sensitivity of each factor, helping to assess which factors play the most significant role in temperature changes. This quantitative assessment can provide more precise control strategies for urban temperature management, especially in local temperature regulation, offering decision support for policymakers. Spatial correlation analysis helps identify the spatial characteristics of locally correlated data. Urban temperature is influenced by factors such as the surrounding environment, building structure, and green spaces. Spatial correlation analysis reveals which areas or structures exhibit strong interrelationships in temperature changes, providing a basis for urban planning and environmental optimization, and helping to design more climate-adaptive urban spatial layouts. Multivariate coupling analysis delves into the interactions between multiple factors, revealing the complexity of temperature changes under the influence of different factors. For example, factors such as building density, wind speed, and green space coverage work together to produce nonlinear effects on temperature changes. This coupled analysis can identify potential risks under the combined influence of multiple factors, optimizing urban climate management strategies. Weighting calculations determine the weights of different factors in heat exposure risk assessment. This process allows the model to adjust its predictions of heat exposure risk based on the degree of influence of different factors, thereby enhancing the model's predictive ability. Simultaneously, based on this weighted data, the identification of critical values ​​helps determine which areas or situations will experience high heat exposure risk, providing a scientific basis for urban climate management. The extraction of critical value data, especially the highest value, represents the potential heat exposure risk threshold. Identifying this threshold allows for the determination of which regions, time periods, or environmental conditions will face the most severe heat exposure risk in the city. This provides an early warning mechanism for governments and relevant departments, helping them to take preventative measures, such as adjusting urban building layouts and improving urban wind corridors, thereby reducing heat exposure levels in high-risk areas.

[0058] Preferably, step S4 includes the following steps:

[0059] Step S41: Import the processed original urban morphology data and urban meteorological data into the urban temperature inference model to perform Monte Carlo simulation and generate temperature simulation samples;

[0060] Step S42: Classify the simulated temperature samples for high-temperature warnings based on the heat exposure risk threshold, and generate high-temperature warning classification data;

[0061] Step S43: Identify risk hotspots in the city using high temperature warning classification data, and generate urban heat exposure risk area data.

[0062] This invention imports processed urban morphology and meteorological data into an urban temperature inference model, using Monte Carlo simulation to generate temperature simulation samples. This process takes into account various uncertainties and simulates multiple urban temperature scenarios through extensive random sampling. The Monte Carlo simulation method effectively improves the reliability of temperature prediction results and provides a more comprehensive data foundation for subsequent risk assessment. High-temperature warning classification data is generated by classifying the temperature simulation samples according to heat exposure risk thresholds. This classification clarifies which areas, time periods, or conditions exceed predetermined heat exposure risk thresholds. This process not only improves the accuracy of warnings but also enables early detection of potential risks from high temperatures, providing data support for subsequent preventative measures. Using the high-temperature warning classification data, risk hotspot identification is conducted, generating urban heat exposure risk area data. This analysis identifies which urban areas are more susceptible to heat exposure risks under extreme weather conditions. This process helps urban planners, policymakers, and public safety departments accurately pinpoint high-risk areas and implement targeted countermeasures. By identifying urban heat exposure risk areas, governments and relevant departments can develop more precise response strategies. For example, in high-risk areas, strategies such as increasing shading facilities and adjusting urban building layouts can be prioritized. These measures can effectively reduce the health threats posed by high temperatures to residents and mitigate the negative impacts of heat exposure. Classifying high-temperature warnings and identifying heat exposure risk areas helps improve a city's adaptability to extreme weather. By accurately identifying high-risk areas, cities can adopt more targeted and effective adaptation measures when facing climate change, such as modifying high-temperature hotspots and designing for climate comfort, thereby enhancing overall climate resilience. Attached Figure Description

[0063] Figure 1 This is a schematic diagram illustrating the steps of a method for urban temperature inference and heat exposure risk assessment based on AdaBoost.

[0064] Figure 2 for Figure 1 A detailed flowchart illustrating the implementation steps of step S2.

[0065] Figure 3 for Figure 1 A detailed flowchart illustrating the implementation steps of step S3.

[0066] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0067] The technical method of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0068] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0069] It should be understood that although the terms "first," "second," etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are used merely to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0070] To achieve the above objectives, please refer to Figures 1 to 3 A method for urban temperature inference and heat exposure risk assessment based on AdaBoost, the method comprising the following steps:

[0071] Step S1: Obtain the original urban morphology data and urban meteorological data of the city; perform data preprocessing on the original urban morphology data and urban meteorological data of the city to generate processed original urban morphology data and urban meteorological data; integrate the processed original urban morphology data and urban meteorological data into the model training set and the model test set.

[0072] Step S2: Design an artificial neural network architecture using the AdaBoost algorithm, and use the artificial neural network architecture to train the model on the training set to generate a pre-model for inferring urban temperature; validate the pre-model for inferring urban temperature based on the test set to generate a model for inferring urban temperature.

[0073] Step S3: Perform global sensitivity analysis and local correlation analysis on the parameters in the urban temperature inference model to generate global sensitivity data and local correlation data; confirm the heat exposure risk threshold based on the global sensitivity data and local correlation data to obtain the heat exposure risk threshold;

[0074] Step S4: Import the processed original urban morphology data and urban meteorological data into the urban temperature inference model for Monte Carlo simulation to generate temperature simulation samples; identify risk hotspots in the temperature simulation samples based on the heat exposure risk threshold to generate urban heat exposure risk area data.

[0075] This invention ensures data quality and consistency by acquiring and preprocessing urban morphology and meteorological data, providing a reliable foundation for subsequent model training and prediction. The artificial neural network architecture designed using the AdaBoost algorithm effectively improves the model's prediction accuracy and generalization ability, especially performing exceptionally well when handling complex nonlinear relationships. The prediction model is validated using a test set, ensuring its reliability and stability and avoiding overfitting or underfitting. Global sensitivity analysis and local correlation analysis help identify key parameters in the model, further optimizing model performance and providing a scientific basis for confirming heat exposure risk thresholds. Based on global sensitivity data and local correlation data, heat exposure risk thresholds can be determined more accurately, providing a clear reference standard for urban thermal environment management. This threshold can be used to identify high-risk areas, helping decision-makers formulate targeted response measures. A large number of temperature simulation samples are generated through Monte Carlo simulation, comprehensively covering various temperature change scenarios and enhancing the robustness of the prediction results. Combined with heat exposure risk thresholds, heat exposure risk areas in cities can be effectively identified, providing important references for urban planning, emergency management, and public health. This method is not only applicable to current urban thermal environment analysis, but can also be applied to heat exposure risk assessment under future climate change scenarios by updating data and models. Its modular design (such as data preprocessing, model training, and simulation analysis) makes it highly flexible and scalable, allowing for adjustments and optimizations to meet the needs of different cities. Therefore, this invention improves the accuracy and reliability of urban temperature inference and heat exposure risk assessment by fusing multi-source data, applying advanced machine learning algorithms, and enhancing sensitivity analysis and Monte Carlo simulation.

[0076] In this embodiment of the invention, reference is made to Figure 1 The diagram shown illustrates the steps of an AdaBoost-based method for urban temperature inference and heat exposure risk assessment according to the present invention. In this example, the AdaBoost-based method for urban temperature inference and heat exposure risk assessment includes the following steps:

[0077] Step S1: Obtain the original urban morphology data and urban meteorological data of the city; perform data preprocessing on the original urban morphology data and urban meteorological data of the city to generate processed original urban morphology data and urban meteorological data; integrate the processed original urban morphology data and urban meteorological data into the model training set and the model test set.

[0078] In this embodiment of the invention, urban morphology data, including building distribution, road grid, green space coverage, and land use types, is acquired. This data can be obtained through remote sensing imagery (such as satellite images) or a Geographic Information System (GIS). Urban meteorological data, including real-time and historical meteorological data such as temperature, humidity, wind speed, and precipitation, is acquired through meteorological stations or online meteorological data platforms. The acquired urban morphology data is cleaned, removing missing or invalid data; the data format is standardized, and data from different sources is standardized (e.g., by unifying coordinate systems and adjusting resolution); different geographic units are clustered to ensure that the data for each urban area reflects the actual situation. The meteorological data is denoised, missing values ​​are filled (e.g., using interpolation to fill missing data), and variables such as temperature are standardized; data smoothing is performed based on temporal and spatial distribution to eliminate the impact of random fluctuations. The processed urban morphology data and meteorological data are then integrated. Urban morphology data and meteorological data for the same time period and spatial area can be paired using spatial matching or temporal synchronization to form data pairs. Based on the time series and spatial partitioning of the data, the integrated data is divided into a training set and a test set for the model according to a certain ratio (e.g., 70% for the training set and 30% for the test set). The training set is used to train the model, and the test set is used to verify the accuracy of the model.

[0079] Step S2: Design an artificial neural network architecture using the AdaBoost algorithm, and use the artificial neural network architecture to train the model on the training set to generate a pre-model for inferring urban temperature; validate the pre-model for inferring urban temperature based on the test set to generate a model for inferring urban temperature.

[0080] In this embodiment of the invention, a suitable artificial neural network architecture, such as a fully connected feedforward neural network (MLP) or a convolutional neural network (CNN), is selected based on the nature of the problem, especially if the data contains spatial structure information. The number of neurons in the input layer should match the number of features in the model training set, for example, including urban morphology data and meteorological data (such as temperature, humidity, wind speed, etc.). The output layer has one neuron (representing the predicted urban temperature value), or multiple outputs can be designed according to specific task requirements. One or more hidden layers are designed, and the number of neurons in each layer can be optimized through cross-validation. The nonlinear activation function of the hidden layer (such as ReLU, Sigmoid, or Tanh) helps to capture complex relationships in the data. The AdaBoost algorithm is used to enhance the performance of the neural network model. AdaBoost is an ensemble learning algorithm that builds a strong prediction model by weighted combination of multiple weak classifiers (in this case, neural network models). AdaBoost dynamically adjusts the weights of samples during training, making subsequent learning processes pay more attention to samples that were previously misclassified, thus improving the model's ability to handle complex situations and reducing overfitting. Multiple neural network-based sub-models are trained, and the outputs of these sub-models are weighted and fused to obtain the final prediction result. The processed model training set is input into the neural network for training. During training, the backpropagation algorithm is used to optimize the network parameters, adjusting weights and biases by minimizing the loss function (such as mean squared error, MSE). During training, hyperparameters such as learning rate, batch size, and regularization can be adjusted to optimize the neural network training process and prevent overfitting or underfitting. After each training round, the AdaBoost algorithm is used to update the sample weights, allowing the next training round to focus more on samples misclassified in the previous round. The model test set is input into the trained urban temperature inference pre-model for prediction, and the results are compared with actual observation data. Appropriate evaluation metrics (such as mean squared error, mean absolute error, MAE, and coefficient of determination, R0) are used. 2 The model's predictive performance was validated using methods such as cross-validation. If performance was unsatisfactory, optimization could be achieved by adjusting the model structure or hyperparameters during training. The model's generalization ability was further verified using methods such as cross-validation to avoid performance degradation on the test set due to overfitting. Through the above training and validation process, an optimized urban temperature inference model was obtained. This model can accurately predict temperature based on input urban morphology and meteorological data, providing a basis for heat exposure risk assessment.

[0081] Step S3: Perform global sensitivity analysis and local correlation analysis on the parameters in the urban temperature inference model to generate global sensitivity data and local correlation data; confirm the heat exposure risk threshold based on the global sensitivity data and local correlation data to obtain the heat exposure risk threshold;

[0082] In this embodiment of the invention, global sensitivity analysis aims to identify the degree of influence of each parameter in the urban temperature inference model on the predicted result (temperature). Therefore, the key parameters in the model are first determined, such as meteorological data (temperature, humidity, wind speed, etc.), urban morphology data (sky view factor, exterior wall area, proportion of hard paving, and ground reflectivity, etc.), and their interaction effects. Commonly used global sensitivity analysis methods include the Sobol method and analysis of variance. The Sobol method based on Monte Carlo simulation can be selected, which evaluates the contribution of each parameter's variation to the output (temperature prediction) by sampling the input parameters. A large number of samples are generated through random sampling techniques (multiple simulations), and the influence of each parameter on the result is calculated by predicting the output through the model. The sensitivity index of each input variable is calculated, such as the variance contribution rate of each parameter and the total variance contribution rate. These indicators help identify key factors that have a significant impact on temperature prediction. Local correlation analysis focuses on the correlation of a specific range or specific input variable in temperature prediction, mainly to identify which factors have a strong influence on temperature inference under specific circumstances (such as a certain climate condition or urban area). Commonly used local correlation analysis methods include Pearson correlation coefficient analysis and mutual information analysis. These methods can be used to assess the linear or nonlinear relationship between a single input feature and temperature prediction results, obtaining the degree of influence of each feature on temperature prediction results in a local context, especially for specific urban areas or meteorological conditions. The heat exposure risk threshold refers to the temperature value exceeding a certain threshold under certain conditions that triggers heat exposure risk and threatens public health. Through global sensitivity analysis and local correlation analysis, key factors related to high-temperature exposure (such as urban building density, temperature, humidity, etc.) are identified, and risk thresholds are determined based on these factors. Combining sensitivity analysis data, a heat exposure risk model is designed, using temperature prediction as input and incorporating urban morphology and meteorological conditions to calculate a heat exposure risk index. This model can be trained using classification methods such as logistic regression and support vector machines. A heat exposure risk threshold is set based on the model's output heat exposure risk index (e.g., when the risk index exceeds a certain value, it indicates the existence of a high-risk heat exposure area). A reasonable threshold can be selected using data-driven methods (such as rules of thumb or statistical analysis). Based on historical data or expert judgment, a critical temperature or risk score is set as the heat exposure risk threshold. By analyzing global and local sensitivity data, heat exposure risk thresholds under different scenarios are determined to ensure the model's adaptability and accuracy. These thresholds are then applied to the model, combined with temperature prediction results, to identify potential high-risk heat exposure areas caused by urban morphology (e.g., high-density building areas) or meteorological conditions (e.g., high humidity, high temperature) leading to the heat island effect. Spatial analysis techniques (e.g., GIS) are used to pinpoint these high-risk areas and generate corresponding risk maps.By combining historical meteorological data and urban planning data, the heat exposure risk level of a specific area can be provided.

[0083] Step S4: Import the processed original urban morphology data and urban meteorological data into the urban temperature inference model for Monte Carlo simulation to generate temperature simulation samples; identify risk hotspots in the temperature simulation samples based on the heat exposure risk threshold to generate urban heat exposure risk area data.

[0084] In this embodiment of the invention, by ensuring that the data preprocessing in step S1 has been completed, the processed urban morphology data (such as building density, green space coverage, road grid, etc.) and urban meteorological data (such as temperature, humidity, wind speed, etc.) are integrated into a standardized dataset. The processed dataset is then input into the urban temperature inference model. This model combines urban morphology and meteorological data to generate a temperature prediction result for the city. Monte Carlo simulation uses random sampling techniques to generate different meteorological and urban morphology scenarios, thereby simulating various temperature changes. For each scenario, the model generates a temperature prediction result. The Monte Carlo method is used to randomly sample uncertainties in meteorological data (such as temperature fluctuations, humidity changes, etc.) and random factors in urban morphology data (such as building materials, urban area characteristics, etc.). Through multiple simulations (e.g., 1000 times or more), a large number of temperature simulation samples are generated. According to a preset probability distribution, meteorological data and urban morphology data are sampled to generate a set of simulation inputs. Each set of simulation inputs is fed into the urban temperature inference model to generate the corresponding temperature prediction value. The temperature values ​​of each simulation are recorded to form a temperature simulation sample set. Based on the heat exposure risk threshold in step S3, the temperature samples obtained from the Monte Carlo simulation are compared with the risk threshold. If the simulated temperature value exceeds the set heat exposure risk threshold, the area is considered a high-risk heat exposure zone. Spatial analysis algorithms (such as hotspot analysis algorithms, K-means clustering, etc.) are used to identify areas in the city with higher temperatures and greater risk, which are caused by factors such as the urban heat island effect, excessive building density, or insufficient green space. GIS technology is used to perform spatial distribution analysis on the simulated temperature samples and create a temperature heat map. Clustering algorithms such as K-means or DBSCAN are used to cluster high-temperature areas and identify heat exposure risk hotspots. Based on the identified risk hotspots, these areas are labeled as "high-risk heat exposure zones," and corresponding urban heat exposure risk area data is generated. This data can be a map, table, or risk level layer. Based on the relationship between temperature values ​​and risk thresholds, the city is divided into different heat exposure risk levels (such as low risk, medium risk, and high risk). Corresponding risk levels are assigned to different areas to form a detailed risk map.

[0085] Preferably, obtaining the original urban morphology data and urban meteorological data of the city includes:

[0086] Obtain the geographical location of the city; use 3D modeling technology to extract the building height, building coverage, exterior wall area, proportion of hard paving, and ground reflectivity of the geographical location of the city to obtain urban morphological characteristic data;

[0087] The sky view factor is calculated by using a fisheye camera to measure the geographical location of the city. The sky view factor is then integrated with the city morphology data to obtain the original city morphology data.

[0088] Reference weather stations were set up based on the original urban morphology data of the city.

[0089] Based on the reference weather station, a mobile weather station is installed at the pedestrian height point, and the mobile weather station is used to collect air temperature, relative humidity, wind speed and solar radiation; thus, urban meteorological data is obtained.

[0090] In this embodiment of the invention, geographic boundary information (latitude and longitude range) of the city is obtained from city maps or open data sources (such as OpenStreetMap) using Geographic Information System (GIS) tools. Combined with satellite imagery or digital maps, the obtained geographic range accurately reflects the actual boundaries of the city. 3D modeling techniques (such as LiDAR scanning or photogrammetry) are used to capture the height data of urban buildings from multiple perspectives. Using information such as building height, building density, and ground cover type (such as green space, roads, buildings, etc.), combined with urban building planning data, the building coverage ratio and exterior wall area are calculated. These morphological feature data are obtained using 3D reconstruction data through computer graphics technology. A fisheye camera is used to photograph the urban geographic range, acquiring large-scale sky image data. Image processing techniques (such as image segmentation and edge detection) are used to extract the visible area of ​​the sky from the images captured by the fisheye camera. By calculating the degree of occlusion of the city skyline and the visible range of the sky, a sky visibility factor is obtained. This factor is typically represented by a value between 0 and 1, where 1 indicates that the sky is completely unobstructed and 0 indicates that the sky is completely obstructed. The calculated urban morphological characteristics data (such as building height, coverage, and exterior wall area) and sky view factors are integrated. This data is stored using a database or data structure to ensure it can be queried and analyzed as needed. The data structure can be a multi-dimensional spatial dataset, with each region or block storing the building morphology and sky view factors for that region. Based on the city's geographical location, climate characteristics, and morphological data, representative areas are selected as locations for reference weather stations. Weather stations are installed at the selected locations, typically including measuring equipment for temperature, humidity, wind speed, and radiation. The reference weather station equipment is ensured to be highly accurate and stable, serving as a calibration standard for subsequent mobile weather stations. Based on the city's spatial layout and pedestrian activity areas, multiple pedestrian activity points are selected to install mobile weather stations, which can typically be mounted on backpacks, portable devices, or mobile public facilities. Temperature sensors, humidity sensors, anemometers, and solar radiation sensors are installed on each mobile weather station to collect data in real time. Wireless communication technologies (such as Wi-Fi, LTE, or 5G) are used to transmit the data collected by the mobile weather stations to a central data processing system. The collected data can be displayed or stored in real time for subsequent meteorological analysis, urban climate research, etc. By integrating the original urban morphology data and collected meteorological data, spatial correlation analysis is conducted, such as the impact of buildings on local climate, to generate meteorological models of heat island effect, wind speed distribution, and humidity changes, and to assess the impact of urban planning and design on climate.

[0091] Preferably, the calculation of the sky view factor for the geographical location of the city using a fisheye camera includes:

[0092] The city's sky image is generated by taking pictures of the geographical area of ​​the city using a fisheye camera.

[0093] Orientation analysis is performed on urban sky images to generate urban sky orientation data; based on the urban sky orientation data, urban sky images are partitioned to generate urban sky partition image sets;

[0094] Sky visibility is calculated for the urban sky partition image set to obtain the partition sky visibility; the urban sky partition image is then used to perform skyline contour detection to generate partition sky skyline contour data.

[0095] The sky view factor is calculated for the urban sky zone image set based on the zoned sky visibility and zoned sky skyline contour data. The formula for calculating the sky view factor is as follows:

[0096]

[0097] in, For the sky view factor, This refers to the visible area of ​​the sky in a photograph of the sky hemisphere. This represents the area of ​​the entire sky hemisphere image.

[0098] In this embodiment of the invention, fisheye cameras are installed at different high points or ground areas within the city to ensure that the cameras cover the widest possible area of ​​the sky. At each shooting location, images are taken periodically, capturing sky images from multiple directions. Fisheye cameras, due to their wide-angle field of view, can capture a large area of ​​sky in a single shot, making them suitable for large-scale urban environments. The images should cover different areas of the city, especially the skyline, for subsequent analysis. Image processing techniques, such as edge detection and color analysis, are used to extract different parts of the sky from the fisheye images. Based on the camera's angle and position, azimuth analysis is performed on the images to extract azimuth data for each image region (e.g., sky coverage in the southeast and northwest directions). The azimuth data for each image should include information such as the visible sky range and obstruction status of the area. Based on the sky's azimuth information, each sky image is divided into predetermined regions (e.g., east, south, west, north, or by building distribution), generating multiple partitioned image sets of the city sky. Each partitioned image should include sky information for a specific direction or region. For example, a sky image for one region may include views in the southeast and southwest directions. The image set can be precisely divided using image segmentation algorithms, region growing methods, etc. Image analysis techniques are employed to detect the proportion of visible sky in the sky image. Visibility calculation is typically based on the openness of the skyline and the proportion of obstructions (such as buildings or mountains). For each image segment, its occlusion is analyzed to calculate the sky visibility for each segment. Visibility values ​​can be between 0 and 1, where 0 represents complete occlusion and 1 represents complete visibility. Algorithms for visibility calculation can use image thresholding, edge detection, and region analysis. Skyline detection algorithms (such as edge detection and contour extraction) are used to extract the skyline contour from each image segment. The complexity of the skyline directly affects the calculation of the sky visibility factor. Skyline detection can use gradient-based image processing techniques to detect the boundary between the sky and buildings and further extract contour data. The complexity of the skyline includes the shape, curvature, and continuity of the contour lines. Complex skylines have more intricate contour lines, while simpler ones are relatively simple. The sky visibility factor is calculated based on visibility, occlusion, and skyline complexity, with the following formula: in, For the sky view factor, This refers to the visible area of ​​the sky in a photograph of the sky hemisphere. The area of ​​the entire sky hemisphere image is represented by A. Within each partition, based on the aforementioned A and B values, the sky view factor for each partition is calculated one by one. The sky view factors of all partitions are averaged or weighted to obtain the comprehensive sky view factor for the entire urban area.

[0099] As an example of the present invention, reference is made to Figure 2As shown, step S2 in this example includes:

[0100] Step S21: Based on the random forest permutation importance method, key variables are screened for the processed original urban morphology data and urban meteorological data to generate key variables for urban morphology and key variables for urban temperature;

[0101] Step S22: Introduce model interaction terms; design an artificial neural network architecture based on the AdaBoost algorithm, wherein the artificial neural network architecture includes a base learner and an ensemble strategy;

[0102] Step S23: Determine the feature vectors of the input layer neurons of the artificial neural network architecture based on the key variables of urban morphology, key variables of urban temperature, and model interaction terms;

[0103] Step S24: Use the feature vectors of the input layer neurons to perform a grid search on the artificial neural network architecture to confirm the optimal parameters; train the model on the training set based on the optimal parameters of the artificial neural network architecture to generate a pre-model for inferring urban temperature; validate the pre-model for inferring urban temperature based on the test set to generate a model for inferring urban temperature.

[0104] In this embodiment of the invention, data quality is ensured by cleaning and standardizing the original urban morphology data (such as building height, building coverage, exterior wall area, proportion of hard paving, and ground reflectivity) and urban meteorological data (such as temperature, humidity, wind speed, and solar radiation). Missing values ​​and outliers are handled, and data normalization or standardization is performed. A random forest model is used to train the processed data. This is implemented using machine learning libraries such as sklearn. During training, the random forest calculates an "importance" score for each feature, typically evaluated using information gain-based calculations. Features with higher importance are selected as key variables for urban morphology and urban temperature, generating key variables: key variables for urban morphology include building density, building height distribution, and green space coverage. Key variables for urban temperature include daytime maximum and minimum temperatures, wind speed, and radiation intensity. Based on these key variables, interaction terms are generated. For example, interaction terms between building density and wind speed, or between building height and solar radiation intensity. Interaction terms can be calculated through simple feature multiplication (e.g., variable1 * variable2), which helps the model capture complex nonlinear relationships. Simple weak learners (e.g., decision stumps) are selected as base learners, allowing them to adjust for errors in each iteration. An ensemble strategy based on the AdaBoost algorithm combines multiple weak learners into a single strong learner. In each training round, AdaBoost assigns higher weights to misclassified samples, making subsequent base learners pay more attention to these samples. The basic structure of the ANN is designed, including an input layer, hidden layers (several layers), and an output layer. The hidden layers use the ReLU activation function, and the output layer is set as a regression layer (linear activation function) or a classification layer (softmax activation function) depending on the task. AdaBoost is combined with the ANN, using it as an ensemble strategy to weight the ANN and optimize its training process. Key variables related to urban morphology, temperature, and interaction terms are combined into a single input feature vector. For example, with 5 urban morphology variables, 4 temperature variables, and 3 interaction terms, the dimension of the input layer feature vector is 12. The number of neurons in the input layer corresponds to the dimension of the feature vector, with one neuron for each feature. The input feature vectors are standardized (e.g., Z-score standardization) to ensure that each feature contributes equally to the training of the neural network, preventing certain variables from dominating the model due to different dimensions. A grid search algorithm is used to optimize the hyperparameters of the neural network.Commonly optimized hyperparameters include: the number of hidden layers and neurons per layer, learning rate, regularization parameters (L2, L1 regularization), activation function type (ReLU, sigmoid, etc.), batch size and number of iterations, using cross-validation (e.g., k-fold cross-validation) to ensure robustness of parameter selection, and implementing grid search using sklearn.model_selection.GridSearchCV. Based on the optimal hyperparameter settings obtained from grid search, the ANN is trained using the training set. A suitable loss function (e.g., mean squared error, MSE) is selected for regression task training. Gradient descent optimization is performed using an optimizer (e.g., Adam, SGD). The trained model is validated using a validation set or test set to evaluate its generalization ability. Evaluation metrics typically include RMSE (root mean squared error) and MAE (mean absolute error). If the model performs poorly, the model architecture can be adjusted or hyperparameters can be re-optimized. Based on the test set evaluation results, the final urban temperature inference model is generated, which can be used to predict and infer urban temperature changes.

[0105] Preferably, an artificial neural network architecture is designed based on the AdaBoost algorithm, wherein the artificial neural network architecture includes a base learner and an ensemble strategy including:

[0106] Configure the artificial neural network; configure the hidden layers of the model based on the ReLU activation function, where the formula for the ReLU activation function is shown below:

[0107]

[0108] in As the dependent variable, As the independent variable, The unstandardized coefficients of the independent variable. For the number of parameters, This is the error term; This represents the index range of the independent variables in the regression equation.

[0109] The model output layer is configured based on a linear function, and the hidden layer and the model output layer are integrated as base learners.

[0110] An ensemble strategy is generated by adaptively weighting and cascading base learners using the AdaBoost algorithm.

[0111] In this embodiment of the invention, the number of neurons in the input layer is set according to the data characteristics of the problem (such as urban morphology data, temperature data, etc.). The number of neurons in the input layer is usually consistent with the dimension of the data features. The input data should be standardized to ensure that each feature contributes equally and to avoid certain features dominating during training. The activation function determines the nonlinear transformation in the neural network. ReLU (Rectified Linear Unit) is a widely used activation function in current deep learning. The mathematical expression of ReLU is as follows: f(x) = max(0,x); it maps all negative values ​​to 0, while positive values ​​remain unchanged, which can accelerate network training and avoid the gradient vanishing problem. An appropriate number of hidden layers and the number of neurons per layer are selected according to the complexity of the problem and the amount of data. Grid search or rules of thumb can be used for selection. For example, if the input data is complex, two to three hidden layers can be selected, each with tens to hundreds of neurons. The hidden layers typically use the ReLU activation function so that the model can better fit the nonlinear relationship of the data. Each layer of the neural network contains a linear transformation plus an activation function. When calculating the output of each layer, a linear transformation is first performed: h = Wx + b; where W is the weight matrix, x is the output of the previous layer, b is the bias term, and h is the weighted summation output of this layer. This output is then passed to the ReLU activation function: f(h) = max(0,h); the output of the hidden layer is mapped to the next layer through a weighted summation (W and b). In regression tasks, the output layer typically uses a linear activation function (or identity function), mathematically expressed as: f(x) = x; this means the output layer undergoes no nonlinear transformation, directly outputting the weighted summation result of the hidden layers, suitable for continuous variable prediction tasks (such as temperature prediction). The output layer typically has only one neuron, representing the output of a continuous value (e.g., a predicted city temperature). The output layer is calculated as: y = Wout * hlast + bout; where y is the final prediction result, hlast is the output of the last hidden layer, and Wout and bout are the weights and biases of the output layer. The designed hidden and output layers are then integrated into a single neural network. This neural network receives input features and generates prediction results. The workflow of a base learner: The input layer receives feature data, which undergoes nonlinear transformation through several hidden layers, and finally generates a prediction result through the output layer. The training objective of the base learner is to minimize the loss function (such as mean squared error, MSE), and the network weights are updated through backpropagation. AdaBoost (Adaptive Boosting) is an ensemble learning method that generates a strong learner by weighting multiple weak learners. In each iteration, it assigns greater weight to misclassified samples, thus making the base learner pay more attention to these samples in the next round. In the first round of training, all samples have equal weights. In each round, a new base learner (i.e., a neural network) is trained.For misclassified samples, their weights are increased, causing the next base learner to focus on learning these incorrect samples. Finally, the predictions of all base learners are weighted and averaged to obtain the final prediction. Each base learner (i.e., the neural network) receives a weight based on its performance on the training set. Better-performing base learners receive higher weights. The outputs of the base learners are used to obtain the final prediction through weighted voting or weighted averaging. For regression problems, weighted averaging is typically used. Each AdaBoost iteration trains a new neural network base learner, weighting the samples based on the error from each training round. After training, the predictions of all base learners are weighted and averaged to obtain the final predicted city temperature.

[0112] Preferably, step S23 includes the following steps:

[0113] Step S231: Perform feature vector transformation on the key variables of urban morphology and urban temperature to generate urban morphology feature vector and urban temperature feature vector; perform correlation analysis on the urban morphology feature vector and urban temperature feature vector to generate morphology-temperature correlation data.

[0114] Step S232: Calculate the feature weights of the morphology-temperature correlation data to obtain feature weight data; optimize the feature combination of the morphology-temperature correlation data based on the feature weight data according to the model interaction terms to generate combined feature data;

[0115] Step S233: Perform dimensional transformation based on the combined feature data to obtain transformed dimensional data; initialize neurons using the transformed dimensional data to obtain the feature vector of the input layer neurons.

[0116] In this embodiment of the invention, key variables of urban morphology and key variables of urban temperature are transformed into feature vectors, and correlation analysis is performed to generate morphology-temperature correlation data. These variables include building height, building coverage, and exterior wall area. In this sub-step, these variables are first standardized and then converted into numerical feature vectors. For example, each urban morphology variable is represented as a feature array, where key urban temperature variables include daily maximum temperature, daily minimum temperature, humidity, and wind speed. These temperature variables are processed similarly to obtain temperature feature vectors. Correlation analysis is then performed by calculating the correlation coefficient between the urban morphology feature vector and the temperature feature vector. Common methods include Pearson correlation coefficient. in and These represent morphological variables and temperature variables, respectively. and This is the corresponding mean. This method can reveal the linear relationship between morphology and temperature. The weight of each feature is calculated using a weighted algorithm or an importance-based ranking method (such as feature importance in random forests). Assuming a random forest algorithm is used, its feature importance can be obtained by calculating the information gain: in, It is a feature It's entropy. It is the number of samples. This represents the entropy of each subset. The calculated feature weights indicate the importance of each feature in the model; features with larger weights contribute more to the temperature prediction model. Based on the calculated feature weights and model interaction terms, the morphology-temperature correlation data is optimized. Feature selection techniques (such as L1 regularization or genetic algorithms) can be used to select the optimal feature combination, or further selection can be achieved through cross-validation. If interaction terms (such as the product of morphological and temperature variables) are introduced, the data needs to be combined according to the weights of these interaction terms to generate new features. Finally, combined feature data is obtained, which integrates morphological features, temperature features, and their interaction information, resulting in higher predictive power. Dimensionality reduction algorithms (such as PCA or t-SNE) are used to transform the dimensions of the combined feature data. PCA (Principal Component Analysis) maps the high-dimensional feature space to a low-dimensional space, reducing redundancy and improving computational efficiency. The PCA process determines the principal components by calculating eigenvalues ​​and eigenvectors, selecting the first k principal components to represent the original data. in, It was before The eigenvectors of the principal components These are the corresponding feature values. The dimensionality-reduced data is the transformed dimensionality data, which has lower dimensionality and effectively represents the main information of the original data. Neuron initialization is performed on the transformed dimensionality data. Common initialization methods include: initializing all neuron weights to zero (suitable for small networks but not for deep neural networks); initializing weights according to a normal or uniform distribution to avoid symmetry violation; and initializing based on the number of neurons in the input layer (suitable for the ReLU activation function). ;in This refers to the number of neurons in the input layer. The initialized neuron feature vectors will be used as input to the neural network's input layer to begin the network's training process.

[0117] As an example of the present invention, reference is made to Figure 3 As shown, step S3 in this example includes:

[0118] Step S31: Calculate the full-order sensitivity index of the parameters in the urban temperature inference model based on the Sobol method to generate global sensitivity data;

[0119] Step S32: Perform local correlation analysis on the parameters in the urban temperature inference model to generate local correlation data. The local correlation analysis includes urban morphology factor-temperature nonlinear relationship analysis and reference meteorological parameter-temperature trend analysis.

[0120] Step S33: Confirm the high-risk threshold for heat exposure based on global sensitivity data and local correlation data to obtain the heat exposure risk threshold.

[0121] In this embodiment of the invention, the Sobol method (also known as full-order sensitivity analysis) is a commonly used method for calculating the sensitivity of model parameters, mainly used to quantify the contribution of input variables to the output results. Its core idea is to calculate the overall impact of changes in each input variable on the model output. The method captures the interaction effects between variables through sensitivity calculations of multiple orders. Specifically, the Sobol method generates model running results with different input combinations, and then calculates the sensitivity index based on these results. Based on the urban temperature inference model (including meteorological parameters such as temperature, humidity, and wind speed, as well as urban morphology variables), an initial value or range is assigned to each model parameter. For example, wind speed and building height are input parameters for urban temperature prediction. For each input parameter, the Sobol method is used to perform full-order sensitivity analysis. First, the values ​​of the input variables are sampled multiple times through a global experimental design of the model (such as Latin hypercube sampling). Then, the sensitivity index of each input variable is calculated. Common sensitivity indices include: in, For input parameters Individual sensitivity (i.e.) Contribution to the output results. This represents the total variance of the model. This method allows us to obtain the degree of influence of each input variable on temperature prediction, generating global sensitivity data. For the nonlinear relationship between landscape factors and temperature, it can be modeled using local regression analysis (such as local weighted regression or local multinomial regression). This method estimates regression coefficients within local regions of different data points: in These are regression coefficients estimated within a local region. Landscape factors are key. By performing local regression on the model, the complex nonlinear relationship between landscape factors and temperature can be revealed, generating data on the nonlinear relationship between urban morphology factors and temperature. Wind speed is another important factor affecting urban temperature. Changes in wind speed lead to heat dissipation, thus affecting temperature. When conducting reference meteorological parameter-temperature trend analysis, trend line analysis or piecewise regression can be used to analyze the temperature change trend under different wind speeds. Commonly used regression models are linear regression or polynomial regression: y = α + βv + Ɛ; where y is temperature, v is wind speed, α and β are parameters to be estimated, and Ɛ is the error term. Through this analysis, the trend relationship between wind speed and temperature can be obtained, generating reference meteorological parameter-temperature trend analysis data. The urban morphology factor-temperature nonlinear relationship data and the reference meteorological parameter-temperature trend analysis data are merged to obtain local correlation data, providing a basis for subsequent risk assessment. The global sensitivity data and local correlation data from steps S31 and S32 are integrated to identify key factors affecting temperature changes. Specifically, based on the sensitivity index and local trend analysis results, the factors most sensitive to high-temperature exposure risk are identified. Based on the integrated data, a heat exposure risk assessment is conducted. One approach is to set a critical temperature for heat exposure (e.g., 38°C, 40°C, etc.) as a high-risk threshold. When the temperature exceeds this critical value, certain areas within the city are considered to be in high-risk heat exposure zones. By analyzing global sensitivity data and local correlation data, appropriate risk thresholds can be set based on temperature and other meteorological variables (e.g., wind speed, humidity). Various statistical methods, such as Extreme Value Theory (EVT), can be used to estimate the extreme risks of heat exposure events and confirm the heat exposure risk threshold. Based on model outputs (e.g., number of high-temperature days, heat stress index, etc.), classification can be performed to determine which areas and time periods are in a high-risk state. This process is usually refined in conjunction with urban characteristics (e.g., density, green space coverage). Ultimately, the heat exposure risk threshold is obtained, which is the high-risk temperature value calculated based on multiple factors such as temperature, landscape factors, and wind speed. This threshold can serve as a basis for decision support in urban planning and the design of climate adaptation measures.

[0122] Preferably, step S32 includes the following steps:

[0123] Step S321: Extract the sky view factor, exterior wall area, hard paving ratio, and ground reflectivity from the urban temperature inference model, and discretize the sky view factor, exterior wall area, hard paving ratio, and ground reflectivity to generate the change values ​​of the sky view factor, exterior wall area, hard paving ratio, and ground reflectivity.

[0124] Step S322: Use the urban temperature inference model to predict temperature changes for each selected sky view factor, exterior wall area, ground hard paving ratio, and ground reflectivity, and generate a comprehensive impact factor-temperature change prediction map; perform slope change trend analysis on the comprehensive impact factor-temperature change prediction map to generate urban morphology factor-temperature nonlinear relationship data.

[0125] Step S323: Extract the reference point wind speed, reference point temperature, reference point solar radiation, and reference point humidity from the urban temperature inference model, and perform value change analysis to generate wind speed change values, reference point temperature change values, reference point solar radiation change values, and reference point humidity change values; use the urban temperature inference model to predict temperature changes for each selected wind speed change value, reference point temperature change value, reference point solar radiation change value, and reference point humidity change value, and generate a reference meteorological parameter-temperature change prediction map;

[0126] Step S324: Perform a downward trend analysis on the reference meteorological parameter-temperature change prediction map to generate negative correlation trend data of reference meteorological parameter-temperature change; integrate the urban morphology factor-temperature nonlinear relationship data and the reference meteorological parameter-temperature change negative correlation trend data into local correlation data.

[0127] In this embodiment of the invention, key parameters are extracted from the urban temperature inference model: sky visibility factor, exterior wall area, proportion of hard paving, and ground reflectivity. These parameters affect temperature changes in urban areas, especially in the context of the urban heat island effect. Each parameter is discretized using a discretization method: different values ​​for the sky visibility factor are defined, such as 0.2, 0.5, 0.8, etc., representing the visible sky area in different regions. The area is further divided based on the building's exterior wall design and area, for example: less than 100 m². 2 100-200 m 2 Greater than 200 m 2Based on variations in the proportion of hard paving on the ground, such as 0.1 (less hard paving), 0.3 (medium proportion of hard paving), and 0.6 (high proportion of hard paving), reflectivity values ​​such as low (0.1), medium (0.4), and high (0.8) were set according to the reflectivity of different ground materials. Variation calculations were performed on these discrete values ​​to obtain the variation values ​​of each parameter. The variation values ​​were analyzed to differentiate these parameters under different scenarios, thus providing a basis for subsequent temperature change prediction. An urban temperature inference model was used to predict temperature changes for each discrete value, considering factors such as sky visibility factor, exterior wall area, proportion of hard paving, and ground reflectivity. This model outputs the corresponding temperature changes based on input urban environmental data, meteorological conditions, and building parameters. By simulating different environmental conditions, a comprehensive influence factor-temperature change prediction map was generated, showing the impact of different urban morphology factors on temperature changes. The slope trend of the generated prediction map was analyzed. This analysis reveals the influence of different urban morphology factors (such as sky visibility, exterior wall area, proportion of hard paving, and ground reflectivity) on temperature changes, thereby generating nonlinear relationship data between urban morphology factors and temperature—that is, analyzing and identifying the nonlinear relationship between urban morphology factors and temperature. Key meteorological parameters, such as wind speed, reference point temperature, solar radiation, and humidity, are extracted from the urban temperature inference model. These parameters are important meteorological factors affecting local temperature changes. Variation analysis of these meteorological parameters is conducted to discover their potential impact on temperature changes. For example, an increase in wind speed may lead to a decrease in temperature, while a change in solar radiation intensity may lead to an increase in temperature. Variation values ​​for each parameter are generated, such as wind speed from 5 m / s to 10 m / s, temperature from 20°C to 25°C, and solar radiation from 300 W / m² to 10 m / s. 2 Up to 600 W / m 2 Changes in meteorological parameters, such as temperature, are used to predict temperature changes using an urban temperature inference model, generating a reference meteorological parameter-temperature change prediction map. This map reflects the trend of the impact of meteorological parameter changes on temperature. Trend analysis is performed on the generated reference meteorological parameter-temperature change prediction map, especially on the parts of temperature change showing a downward trend. The analysis results help identify which meteorological parameters (such as wind speed, humidity, etc.) show a negative correlation with temperature changes. Negative correlation trend data of reference meteorological parameters-temperature changes is generated, which reveals how changes in meteorological parameters have a negative correlation effect with temperature changes (e.g., increased wind speed leads to decreased temperature, increased humidity leads to decreased temperature, etc.). The urban morphology factor-temperature nonlinear relationship data is integrated with the negative correlation trend data of reference meteorological parameters-temperature changes to generate local correlation data.

[0128] Preferably, step S33 includes the following steps:

[0129] Step S331: Identify sensitive factors in the global sensitivity data to obtain sensitive factor data; perform parameter response analysis on the global sensitivity data based on the sensitive factor data to obtain response feature data;

[0130] Step S332: Perform sensitivity quantification calculation on the response feature data to obtain quantitative index data; perform spatial correlation analysis on the local correlation data to obtain correlation feature data; perform multivariate coupling analysis on the response feature data and correlation feature data to obtain coupling effect data;

[0131] Step S333: Calculate the influence weights based on the coupling effect data to obtain weight data; use the weight data to identify the critical values ​​of the coupling effect data to obtain critical value data, and extract the highest value of the critical value data as the heat exposure risk threshold.

[0132] In this embodiment of the invention, global sensitivity data is analyzed using the Sobol method to identify the most significant sensitive factors affecting temperature changes. These sensitive factors include building density, sky view factor, and wind speed. Sensitivity analysis techniques, such as variance-based sensitivity analysis and global sensitivity analysis, are employed to quantitatively assess the contribution of each factor. Parametric response analysis is performed on the identified sensitive factors, which involves exploring how changes in these sensitive factors affect the prediction results in the urban temperature inference model. By analyzing the impact of factor changes on the model output (temperature prediction), response characteristic data is obtained, revealing the sensitivity and degree of influence of each sensitive factor on temperature changes. The degree of influence of each sensitive factor on temperature changes is quantified using sensitivity measures (e.g., Sobol index, variance contribution). This step generates quantitative index data reflecting the contribution of each factor in the global sensitivity data. Linear regression, nonlinear fitting, and other methods can be used to quantify the relationship between response characteristics and temperature changes. Spatial correlation analysis is performed on locally correlated data, using Geographic Information System (GIS) methods or spatial autocorrelation analysis to explore the relationship between urban temperature and spatial distribution. For example, Moran's I or Getis-Ord GI* statistics can be used to analyze local correlation data, assess the spatial relationships between landscape factors and temperature in different regions, and generate correlation feature data. Multivariate coupling analysis can then be performed using response feature data and spatial correlation data. Principal component analysis (PCA), canonical correlation analysis (CCA), and other methods can be used to reduce and fuse multidimensional data to obtain coupling effect data between factors. Coupling effect data reveals how different factors (such as landscape factors, temperature, wind speed, etc.) interact spatially and temporally, thus affecting heat exposure levels. Based on the coupling effect data, the weights of each factor and their combined effects in heat exposure are calculated. Common methods include entropy weighting and analytic hierarchy process (AHP), which can help determine the relative importance of each factor in heat exposure risk. By calculating the weights, the influence of each factor on urban temperature changes is obtained, and they are weighted according to their impact on heat exposure. Based on the influence weight data, critical values ​​are identified for the coupling effect data. A critical value refers to the value exceeding which, under specific environmental and climatic conditions, a higher risk of heat exposure is triggered. Threshold analysis methods, such as optimal classification and ROC curve analysis, can be used to determine the critical values ​​for heat exposure risk. The critical value identification process involves modeling the relationship between heat exposure risk and influencing factors to identify key thresholds affecting temperature changes. These critical values ​​can be identified based on historical temperature data, urban heat island effect data, and simulation results. Based on the critical value identification results, the highest value in the critical value data is extracted as the final heat exposure risk threshold.This threshold indicates that once a certain area of ​​the city reaches or exceeds this value, it will face a high risk of heat exposure.

[0133] Preferably, step S4 includes the following steps:

[0134] Step S41: Import the processed original urban morphology data and urban meteorological data into the urban temperature inference model to perform Monte Carlo simulation and generate temperature simulation samples;

[0135] Step S42: Classify the simulated temperature samples for high-temperature warnings based on the heat exposure risk threshold, and generate high-temperature warning classification data;

[0136] Step S43: Identify risk hotspots in the city using high temperature warning classification data, and generate urban heat exposure risk area data.

[0137] In this embodiment of the invention, processed original urban morphology data (such as building height, building coverage, exterior wall area, proportion of hard paving, and ground reflectivity) and urban meteorological data (such as temperature, humidity, and wind speed) are input into a pre-trained urban temperature inference model. It is ensured that the format of the input data matches the model's requirements (e.g., standardization, normalization). The Monte Carlo method is a numerical method for estimating solutions to complex problems through random sampling. Here, Monte Carlo simulation is used to perform multiple random simulations on the temperature inference model, generating a temperature sample for each simulation based on different urban morphology and meteorological conditions. During the simulation, the input data can be randomly perturbed (e.g., by weather changes, building changes, etc.) to obtain temperature distributions under different scenarios. Each simulation step generates a temperature sample, resulting in a series of simulated temperature samples. Based on the Monte Carlo simulation process, multiple simulated temperature samples are generated, reflecting temperature changes under different urban morphologies and meteorological conditions. Through these simulated temperature samples, the temperature distribution range can be obtained, and the trend of urban temperature changes under different conditions can be analyzed. A critical temperature value is set based on the heat exposure risk threshold from the previous step (e.g., the maximum temperature threshold obtained through analysis of global sensitivity data and local correlation data). If the temperature of some simulated temperature samples exceeds this threshold, then the area has a high-temperature exposure risk. Each simulated temperature sample is compared with the heat exposure risk threshold. If the temperature sample exceeds the threshold, it is marked as a high-temperature risk area; otherwise, it is marked as a normal temperature area. All simulated temperature samples are classified to generate high-temperature warning classification data, which marks whether each area has a high-temperature exposure risk. Different classification methods can be used, such as binary classification (high temperature / non-high temperature) or multi-level classification (different levels of high-temperature risk), such as low, medium, and high heat exposure risk. Spatial analysis methods (such as spatial clustering, heat map analysis, hotspot analysis, etc.) are used to process the high-temperature warning classification data to identify multiple areas with excessively high temperatures and high heat exposure risks. K-means clustering, DBSCAN density clustering, and other algorithms can be used to perform cluster analysis on high-temperature areas to find heat exposure risk hotspots. High-risk areas for heat exposure are marked on city maps. These areas are typically characterized by concentrated high temperatures, lack of green space, or low wind speeds, specifically areas where the urban heat island effect is concentrated. Spatial data visualization can be performed using GIS software (such as ArcGIS or QGIS) to visually display the locations of risk hotspots. All identified high-temperature risk areas are integrated to form a detailed urban heat exposure risk area database, including the spatial coordinates, area, and risk level of each high-risk area. Based on the risk classification results, different response strategies can be developed for each area (e.g., high-temperature emergency response, urban greening, and the construction of shading facilities).

[0138] Therefore, the embodiments should be considered as exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of the equivalents of the application are intended to be included within the invention.

[0139] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A method for urban temperature inference and heat exposure risk assessment based on AdaBoost, characterized in that, Includes the following steps: Step S1: Obtain the original urban morphology data and urban meteorological data of the city; perform data preprocessing on the original urban morphology data and urban meteorological data of the city to generate processed original urban morphology data and urban meteorological data; integrate the processed original urban morphology data and urban meteorological data into the model training set and the model test set. Step S2: Design an artificial neural network architecture using the AdaBoost algorithm, and use the artificial neural network architecture to train the model on the training set to generate a pre-model for inferring urban temperature; validate the pre-model for inferring urban temperature based on the test set to generate a model for inferring urban temperature. Step S3: Perform global sensitivity analysis and local correlation analysis on the parameters in the urban temperature inference model to generate global sensitivity data and local correlation data; The heat exposure risk threshold is determined based on global sensitivity data and local correlation data; wherein, step S3 includes the following steps: Step S31: Calculate the full-order sensitivity index of the parameters in the urban temperature inference model based on the Sobol method to generate global sensitivity data; Step S32: Perform local correlation analysis on the parameters in the urban temperature inference model to generate local correlation data. The local correlation analysis includes urban morphology factor-temperature nonlinear relationship analysis and reference meteorological parameter-temperature trend analysis. Step S33: Confirm the high-risk threshold for heat exposure based on global sensitivity data and local correlation data to obtain the heat exposure risk threshold; wherein, step S33 includes the following steps: Step S331: Identify sensitive factors in the global sensitivity data to obtain sensitive factor data; perform parameter response analysis on the global sensitivity data based on the sensitive factor data to obtain response feature data; Step S332: Perform sensitivity quantification calculation on the response feature data to obtain quantitative index data; perform spatial correlation analysis on the local correlation data to obtain correlation feature data; perform multivariate coupling analysis on the response feature data and correlation feature data to obtain coupling effect data; Step S333: Calculate the influence weights based on the coupling effect data to obtain weight data; use the weight data to identify the critical values ​​of the coupling effect data to obtain critical value data, and extract the highest value of the critical value data as the heat exposure risk threshold. Step S4: Import the processed original urban morphology data and urban meteorological data into the urban temperature inference model for Monte Carlo simulation to generate temperature simulation samples; identify risk hotspots in the temperature simulation samples based on the heat exposure risk threshold to generate urban heat exposure risk area data.

2. The method for urban temperature inference and heat exposure risk assessment according to claim 1, characterized in that, Obtaining the original urban morphology data and urban meteorological data of the city includes: Obtain the geographical location of the city; use 3D modeling technology to extract the building height, building coverage, exterior wall area, proportion of hard paving, and ground reflectivity of the geographical location of the city to obtain urban morphological characteristic data; The sky view factor is calculated by using a fisheye camera to measure the geographical location of the city. The sky view factor is then integrated with the city morphology data to obtain the original city morphology data. Reference weather stations were set up based on the original urban morphology data of the city. Based on the reference weather station, a mobile weather station is installed at the pedestrian height point, and the mobile weather station is used to collect air temperature, relative humidity, wind speed and solar radiation; thus, urban meteorological data is obtained.

3. The method for urban temperature inference and heat exposure risk assessment according to claim 2, characterized in that, Calculating the sky view factor for the geographical location of a city using a fisheye camera includes: The city's sky image is generated by taking pictures of the geographical area of ​​the city using a fisheye camera. Orientation analysis is performed on urban sky images to generate urban sky orientation data; based on the urban sky orientation data, urban sky images are partitioned to generate urban sky partition image sets; Sky visibility is calculated for the urban sky partition image set to obtain the partition sky visibility; the urban sky partition image is then used to perform skyline contour detection to generate partition sky skyline contour data. The sky view factor is calculated for the urban sky zone image set based on the zoned sky visibility and zoned sky skyline contour data. The formula for calculating the sky view factor is as follows: ； in, For the sky view factor, This refers to the visible area of ​​the sky in a photograph of the sky hemisphere. This represents the area of ​​the entire sky hemisphere image.

4. The method for urban temperature inference and heat exposure risk assessment according to claim 1, characterized in that, Step S2 includes the following steps: Step S21: Based on the random forest permutation importance method, key variables are screened for the processed original urban morphology data and urban meteorological data to generate key variables for urban morphology and key variables for urban temperature; Step S22: Introduce model interaction terms; design an artificial neural network architecture based on the AdaBoost algorithm, wherein the artificial neural network architecture includes a base learner and an ensemble strategy; Step S23: Determine the feature vectors of the input layer neurons of the artificial neural network architecture based on the key variables of urban morphology, key variables of urban temperature, and model interaction terms; Step S24: Use the feature vectors of the input layer neurons to perform a grid search on the artificial neural network architecture to confirm the optimal parameters; train the model on the training set based on the optimal parameters of the artificial neural network architecture to generate a pre-model for inferring urban temperature; validate the pre-model for inferring urban temperature based on the test set to generate a model for inferring urban temperature.

5. The method for urban temperature inference and heat exposure risk assessment according to claim 4, characterized in that, Step S23 includes the following steps: Step S231: Perform feature vector transformation on the key variables of urban morphology and urban temperature to generate urban morphology feature vector and urban temperature feature vector; perform correlation analysis on the urban morphology feature vector and urban temperature feature vector to generate morphology-temperature correlation data. Step S232: Calculate the feature weights of the morphology-temperature correlation data to obtain feature weight data; optimize the feature combination of the morphology-temperature correlation data based on the feature weight data according to the model interaction terms to generate combined feature data; Step S233: Perform dimensional transformation based on the combined feature data to obtain transformed dimensional data; initialize neurons using the transformed dimensional data to obtain the feature vector of the input layer neurons.

6. The method for urban temperature inference and heat exposure risk assessment according to claim 1, characterized in that, Step S32 includes the following steps: Step S321: Extract the sky view factor, exterior wall area, hard paving ratio, and ground reflectivity from the urban temperature inference model, and discretize the sky view factor, exterior wall area, hard paving ratio, and ground reflectivity to generate the change values ​​of the sky view factor, exterior wall area, hard paving ratio, and ground reflectivity. Step S322: Use the urban temperature inference model to predict temperature changes for each selected sky view factor, exterior wall area, ground hard paving ratio, and ground reflectivity, and generate a comprehensive impact factor-temperature change prediction map; perform slope change trend analysis on the comprehensive impact factor-temperature change prediction map to generate urban morphology factor-temperature nonlinear relationship data. Step S323: Extract the reference point wind speed, reference point temperature, reference point solar radiation, and reference point humidity from the urban temperature inference model, and perform value change analysis to generate wind speed change values, reference point temperature change values, reference point solar radiation change values, and reference point humidity change values; use the urban temperature inference model to predict temperature changes for each selected wind speed change value, reference point temperature change value, reference point solar radiation change value, and reference point humidity change value, and generate a reference meteorological parameter-temperature change prediction map; Step S324: Perform a downward trend analysis on the reference meteorological parameter-temperature change prediction map to generate negative correlation trend data of reference meteorological parameter-temperature change; integrate the urban morphology factor-temperature nonlinear relationship data and the reference meteorological parameter-temperature change negative correlation trend data into local correlation data.

7. The method for urban temperature inference and heat exposure risk assessment according to claim 1, characterized in that, Step S4 includes the following steps: Step S41: Import the processed original urban morphology data and urban meteorological data into the urban temperature inference model to perform Monte Carlo simulation and generate temperature simulation samples; Step S42: Classify the simulated temperature samples for high-temperature warnings based on the heat exposure risk threshold, and generate high-temperature warning classification data; Step S43: Identify risk hotspots in the city using high temperature warning classification data, and generate urban heat exposure risk area data.

Images