Highway pollution reduction and carbon reduction scheme screening method based on unmanned aerial vehicle stereo monitoring and ai collaborative decision

Abstract

This invention discloses a method for selecting pollution reduction and carbon reduction schemes for elevated roads based on UAV three-dimensional monitoring and AI collaborative decision-making. The method constructs a high-resolution pollution field database through UAV three-dimensional surveys, calibrates a CFD model using measured data, and trains a multi-model ensemble surrogate model. It performs structured analysis and multiple verifications on user-input design objectives and constraints to generate a candidate scheme set. The surrogate model enables rapid evaluation of the pollution reduction and carbon reduction performance of candidate schemes, and combined with hard constraint screening and multi-objective optimization, outputs recommended schemes, and visualizes and dynamically updates the results. This invention achieves automatic generation, rapid evaluation, and intelligent selection of pollution reduction and carbon reduction schemes for elevated roads, providing decision support for the green and low-carbon construction of elevated roads.

Description

Technical Field

[0001] This invention relates to the field of urban road planning technology, and in particular to a method for screening pollution reduction and carbon reduction schemes for elevated roads based on UAV three-dimensional monitoring and AI collaborative decision-making. Background Technology

[0002] In recent years, with the continuous growth of urban motor vehicle ownership and the increasing pressure on the road network in central urban areas, elevated roads have become an important infrastructure for alleviating ground congestion and improving traffic conversion efficiency. At the same time, elevated bridges alter the spatial location of pollution sources, local wind field organization, and the pollution characteristics of surrounding building interfaces. Especially in complex scenarios such as "high-rise buildings on both sides" or "high-rise buildings on one side and open space on the other," they easily form capping effects, backflow vortices, and localized retention zones, leading to abnormal accumulation of pollutants such as fine particulate matter and black carbon under the bridge, on the bridge sides, and near the window height of buildings facing the street. Existing research mostly focuses on the analysis of pollution diffusion mechanisms, with less emphasis on further addressing the engineering decision-making question of which elevated road construction or renovation methods should be adopted under given pollution reduction and carbon reduction targets and engineering boundary conditions. Traditional monitoring methods mainly rely on ground-based fixed stations, mobile monitoring, or a small number of vertical observation devices, making it difficult to reliably acquire high spatiotemporal resolution three-dimensional pollution field data around elevated roads. Although computational fluid dynamics (CFD) models can provide high-resolution flow field and concentration field information, their computational overhead is large, making it difficult to support rapid comparison and selection of a large number of candidate schemes during the planning and design phase. Existing machine learning or surrogate models mostly focus on concentration estimation and lack a scheme screening mechanism coupled with engineering constraints, spatial constraints, and carbon emission constraints.

[0003] Therefore, it is necessary to propose a new technical solution that organically combines UAV three-dimensional monitoring, multi-source data fusion, field measurement, calibration and simulation, rapid evaluation and solution selection, so as to provide feasible decision support for the green and low-carbon construction of elevated roads. Summary of the Invention

[0004] Based on the above, this invention aims to propose a method for screening pollution reduction and carbon reduction schemes for elevated roads based on UAV three-dimensional monitoring and AI collaborative decision-making. It targets three typical elevated road scenarios: high-rise buildings on both sides, high-rise buildings on one side and open space on the other, and open space on both sides. Through UAV three-dimensional monitoring, pollution field database construction, CFD model calibration, high-fidelity simulation sample library construction, and surrogate model training, a rapid assessment basis for pollution reduction and carbon reduction in different typical scenarios is formed. Based on this, according to the target area, constraints, and preference input by the user, corresponding elevated road construction or renovation schemes are generated and screened.

[0005] To achieve the above-mentioned technical objectives, the technical solution adopted by this invention is as follows: This invention proposes a method for screening pollution reduction and carbon reduction schemes for elevated roads based on UAV three-dimensional monitoring and AI collaborative decision-making, including the following steps: S1: The target elevated road area is surveyed in three dimensions using multiple monitoring modules carried by the UAV to obtain multi-source observation data, which is then integrated to construct a high-resolution pollution field database. The CFD model is calibrated based on the measured data to generate a high-fidelity simulation sample library and train a multi-model integrated surrogate model. S2: Receive user input of design goals, constraints, and preference preferences, perform structured analysis and transformation, and combine them with a pre-built domain knowledge base, regulatory clause base, engineering capability base, carbon emission factor base, and historical case base to form a structured set of goals and constraints required for candidate solution generation and screening; S3: The structured target constraint set is sequentially checked through the multi-dimensional verification submodule. Parameters with conflicts or missing parameters are corrected and feedback is provided until a solution that has passed all verifications is generated and the task input for filtering is formed. S4: Based on the task input verified, construct a candidate solution search space, perform combined search and optimization on adjustable design variables, and generate and pre-screen a set of candidate solutions that meet the basic constraints. S5: Use the proxy model obtained in step S1 to evaluate the candidate solutions, call the CFD model to verify the solutions that meet the preset conditions, and combine interpretable artificial intelligence to quantify the contribution of each design variable to form a multi-index evaluation result matrix. S6: Based on preset index thresholds and engineering constraints, perform hard constraint screening and multi-objective optimization on candidate solutions, and output recommended solutions, alternative sequences and supporting design strategies. S7: Generate visualized solution results and establish a traceable audit chain throughout the entire process to enable dynamic re-evaluation and rolling updates of the solution.

[0006] In some embodiments, step S1 specifically includes the following steps: S1.1: Use UAVs equipped with multiple monitoring modules to conduct multi-height, cross-sectional three-dimensional spatial surveys of the target elevated road area, collect particulate matter concentration, local meteorological parameters, three-dimensional point cloud, on-site images and UAV attitude trajectory data, and verify the data in conjunction with ground auxiliary monitoring equipment when necessary; S1.2: Synchronously process multi-source observation data according to a unified time reference. Through coordinate transformation, spatial interpolation, point cloud registration and profile reconstruction, map the data from UAVs, ground sensors and static planning data to the three-dimensional raster coordinate system or parameterized scene unit of the target elevated road area. S1.3: Combining the road design drawings, BIM model, GIS base map, land boundary, traffic flow and vehicle emission factor database of the target area, extract the basic features of the elevated roads and surrounding environment of the target area; perform raster assignment, spatial interpolation, attribute association and standardized storage on the mapped pollutant concentration data, micro-meteorological data and three-dimensional spatial morphology data to obtain the multi-dimensional parameters corresponding to each three-dimensional raster unit, thereby forming a high-resolution pollution field database; S1.4: Using high-resolution pollution field data obtained from UAV field measurements as the calibration benchmark, the CFD model of the target elevated road scene is adaptively calibrated; multiple sets of combined sampling are performed on the elevated road geometric parameters, traffic operation parameters, wind field parameters and surrounding building parameters through the Latin hypercube sampling method; based on the calibrated CFD model, the corresponding pollution diffusion results and carbon emission results are generated, and a high-fidelity simulation sample library is constructed. S1.5: The elevated geometric parameters, traffic operation parameters, wind field parameters, surrounding building parameters, and ancillary facility parameters of each group of samples are used as input features, and the corresponding pollution diffusion indicators, carbon emission indicators, ventilation performance indicators, or concentration distribution results of virtual monitoring points in the target area are used as output labels to construct a training sample set for the surrogate model; then the training sample set is normalized and divided into a training subset and a validation subset, and a support vector regression model, an XGBoost model, and a deep neural network with attention mechanism are trained respectively to obtain multiple base learners; Then, the prediction results of each base learner are used as secondary inputs, and a meta-learner is used to weight and fuse them to obtain a multi-model ensemble proxy model for rapid evaluation of candidate solutions.

[0007] In some embodiments, the static planning data mentioned in step S1.2 includes at least one or more of the following: road design drawings, as-built drawings, BIM models, GIS base maps, topographic data, land boundaries, regulatory detailed planning data, building outlines and height data, road red line data, bridge structure dimensions data, municipal ancillary facility layout data, and functional zoning data of the space under the bridge. The parameterized scene unit is a set of structured parameters extracted according to the spatial morphology of roads, bridges, buildings, and the space under bridges.

[0008] In some embodiments, step S2 specifically includes the following steps: S2.1: Receive user input regarding design goals, constraints, and preference criteria; S2.2: Perform structured extraction, unit normalization, noun disambiguation, and consistency verification on the input design objectives, constraints, and optimization preferences, and transform them into calculable target parameters, constraint parameters, evaluation indicators, and design variable boundaries; S2.3: Based on the pre-built domain knowledge base, regulatory clause base, engineering capability base, carbon emission factor base and historical case base, and combined with the preset rule templates and constraint matching rules, the target parameters, constraint parameters, evaluation indicators and design variable boundaries are completed, verified and organized, and the weight allocation of multiple objectives is completed according to the user's preference to form a set of scheme selection conditions; S2.4: When there are conflicts among input objectives, key constraints are missing, or variable boundaries are unclear, generate items to be confirmed and prompt manual supplementation or revision until a structured task input that can be used for subsequent candidate solution generation, pollution reduction and carbon reduction assessment, and multi-objective selection is formed.

[0009] In some embodiments, the design objectives in step S2.1 include at least one or more of the following: target area, design stage, pollution control objectives, carbon emission control objectives, cost control objectives, under-bridge space utilization objectives, sensitive target protection requirements, and construction implementation objectives; the constraints include at least one or more of the following: airspace requirements, land use boundaries, budget limits, construction period requirements, structural modification restrictions, traffic organization restrictions, and operation and maintenance conditions. In step S2.2, structured extraction involves semantic parsing, parameter slot extraction, and parameter mapping of user input to identify key parameters including target area, typical scene type, building height, building setback distance, bridge structure dimensions, pollution control threshold, carbon emission threshold, and engineering constraints. The design variable boundary refers to the adjustable range, prohibition conditions, and value restrictions set for the main form of the elevated road, the treatment of the space under the bridge, the configuration of ancillary facilities, the control method of surrounding buildings, the greening configuration method, and traffic organization parameters.

[0010] In some embodiments, step S3 specifically includes the following steps: S3.1 Input the structured target constraint set formed in step S2 into the regulatory compliance verification submodule, the engineering feasibility verification submodule, the spatial constraint verification submodule, and the data applicability verification submodule in sequence, and perform rule matching and constraint detection on the target parameters, constraint parameters, evaluation indicators, and design variable boundaries; The aforementioned regulatory compliance verification involves legally verifying road design specifications, clearance requirements, land use boundaries, safety distances, environmental quality targets, and requirements for the use of space under bridges. The data applicability verification involves combining monitoring density, scenario completeness, and model applicability range to mark designs that rely on low-reliability data or exceed the model's applicable conditions. S3.2: When a constraint conflict, missing parameter, unclear boundary, or exceeding the engineering capability is detected, the system generates feedback information including the source of the conflict, the violation condition, the affected variables, and alternative suggestions, and prompts for manual confirmation. S3.3: Re-execute structured updates and constraint checks on the revised objectives and constraints that have been manually confirmed until the structured objective and constraint set passes all checks; if there are irremovable conflicts, retain the conflicting items and limit the scope of variables in the subsequent candidate solution generation process; S3.4: Organize all verified target parameters, constraint parameters, evaluation indicators, and design variable boundaries into inputs for scheme generation and screening tasks.

[0011] In some embodiments, step S4 specifically includes the following steps: S4.1: Based on the verified structured objective constraint set, determine the design variable boundaries, variable value ranges, and prohibitive conditions. Combined with the site conditions extracted from the high-resolution contaminated field database formed in step S1 and the task inputs formed in steps S2 and S3, construct a search space for candidate elevated road construction or renovation schemes. The search space includes at least the main form of the elevated road, the opening method of the space under the bridge, the configuration of the auxiliary barriers, the control method of surrounding buildings, the greening or functional material strategy, and the traffic organization coordination strategy. S4.2: First, read the set of adjustable design variables, the range of variable values, the basic prohibition conditions, and the target weight information; then, according to the preset combination rules, parameterize and encode each design variable, and under the premise of meeting the requirements of clearance, land boundary, budget constraints, structural safety requirements, and sensitive target protection requirements, perform a combination search on the adjustable design variables to generate an initial candidate scheme set; Subsequently, the historical case similarity retrieval module is invoked to extract the variable combination features of high-performing solutions in similar scenarios, and the search space is preferentially sampled or reduced accordingly. Finally, combined with heuristic optimization strategies or preference-guided strategies, the initial candidate solution set is iteratively expanded, filtered, and sorted to obtain multiple candidate solutions that meet the constraints. S4.3: Standardize and code the candidate solutions, representing each solution as a parameter vector or graph structure consisting of geometric variables, environmental variables, material variables, traffic variables, and control variables, and eliminate infeasible solutions based on basic constraints; among which, basic constraints include clearance, construction, land boundary, budget, and operation and maintenance conditions; S4.4: Establish the binding relationship between candidate solutions and target area scenarios to form a set of candidate solutions that have undergone preliminary screening.

[0012] In some embodiments, step S5 specifically includes the following steps: S5.1: Match and merge the parameters of the preliminary screening candidate schemes with the background parameters of typical scenarios in the target area to establish a collaborative assessment input for pollution reduction and carbon reduction for each candidate scheme; The background parameters are composed of surrounding building heights, building setback distances, bridge structure dimensions, traffic operation parameters, local meteorological parameters, background pollution parameters, and spatial morphology parameters corresponding to the identified typical scenarios, either input by the user or extracted from planning data. The collaborative evaluation input includes at least one or more of the following: elevated road main morphology parameters, bridge under-bridge space processing parameters, surrounding building morphology parameters, traffic operation parameters, local meteorological parameters, background pollution parameters, and carbon emission-related parameters. S5.2: Based on the multi-model integrated proxy model obtained in step S1, the pollution transport process and ventilation performance of the candidate schemes are simulated to obtain the pollutant concentration distribution, exposure risk indicators, and ventilation performance related indicators; S5.3: Conduct coordinated quantification of carbon emissions during the construction and operation phases of candidate projects; wherein, carbon emissions during the construction phase shall include at least the carbon emissions of bridge materials, ancillary components, greening facilities and functional materials, and carbon emissions during the operation phase shall include at least the carbon emissions generated by changes in traffic operation, energy consumption of ancillary facilities operation and maintenance activities; S5.4: Use a pre-trained surrogate model to quickly extrapolate multiple candidate schemes, output pollution diffusion indicators, carbon emission indicators, exposure risk indicators, ventilation performance indicators, cost indicators and engineering implementation indicators corresponding to each candidate scheme, and simultaneously provide the model prediction confidence or uncertainty evaluation. S5.5: Based on interpretable artificial intelligence analysis, combined with SHAP value analysis, attention weight analysis, sensitivity analysis or counterfactual interpretation, quantify the degree of influence of each design variable on pollution control effect and carbon emission results, obtain parameter contribution information, and form a multi-index evaluation result matrix.

[0013] In some embodiments, step S6 specifically includes the following steps: S6.1: Receive the target indicator threshold and engineering constraints set by the user, and match them with the multi-indicator evaluation result matrix output in step S5; The target indicator thresholds include at least one or more of the following: pedestrian level pollutant concentration threshold, building window height pollution threshold, bridge underpass area exceeding the standard volume ratio threshold, carbon emission per unit length threshold, unit pollution reduction benefit threshold, and budget ceiling; the engineering constraints include at least one or more of the following: airspace constraints, land use boundary constraints, structural modification restrictions, construction period constraints, and construction accessibility constraints. S6.2: Perform hard constraint screening on candidate solutions, eliminate candidate solutions that do not meet the requirements on any constraint, and obtain a set of feasible solutions that meet the basic requirements; S6.3: Perform multi-objective optimization on the set of feasible solutions, wherein the multi-objective optimization is accomplished by one or more combinations of weight-based comprehensive score ranking, Pareto front non-dominated ranking, and rule priority recommendation; S6.4: Based on the multi-objective optimization results, design strategies are extracted from the set of feasible solutions. These design strategies include not only the final recommended scheme, but also corresponding suggestions on the elevated main structure form, the treatment of the space under the bridge, the control of surrounding buildings, the configuration of green spaces and functional materials, and the coordination of traffic organization.

[0014] In some embodiments, the visualization results in step S7 include at least one or more of the following: pollution diffusion cloud map, exposure heat map, carbon emission comparison map, scheme radar map, and parameter contribution interpretation map; the full-process audit chain is a set of associated records that can trace back the source of demand, the source of scenario data, the configuration of key parameters, the model call relationship, and the output results. The dynamic re-evaluation and rolling update of the scheme are carried out in the following way: when the monitoring data of the target area, traffic conditions, meteorological conditions, scene boundary conditions or user-defined thresholds change, the updated scene database is called, and incremental learning or transfer updates are performed on the proxy model based on the newly added monitoring samples. The candidate scheme evaluation, feasibility screening and multi-objective optimization steps are re-executed to realize the dynamic re-evaluation and rolling update of the scheme.

[0015] By adopting the above technical solution, the present invention has the following beneficial effects compared with the prior art: 1. This invention constructs a complete technical system encompassing "UAV three-dimensional perception - intelligent semantic analysis - intelligent generation of candidate solutions - actual measurement, calibration and simulation - interpretable AI rapid evaluation - target constraint screening and optimization - dynamic learning and updating". It breaks through the limitations of existing technologies that only focus on the analysis of pollution diffusion mechanisms, and realizes a closed-loop transformation from pollution characteristic monitoring to engineering solution decision-making. It can directly provide decision support for the construction, reconstruction, expansion and under-bridge space optimization of elevated roads.

[0016] 2. This invention acquires high spatiotemporal resolution three-dimensional pollution field data around elevated roads through multi-altitude, multi-section three-dimensional surveys conducted by UAVs. Combined with multi-source data fusion, a standardized high-resolution pollution field database is constructed, solving the problem that traditional ground monitoring cannot accurately characterize the three-dimensional accumulation characteristics of pollutants in complex elevated road scenarios. This provides a high-precision measured benchmark for subsequent model calibration and scheme evaluation.

[0017] 3. This invention constructs a rapid evaluation method based on measured and calibrated CFD and a multi-model integrated proxy model. The CFD model is adaptively calibrated using UAV measured data as a benchmark, ensuring the authenticity of the simulation results. The multi-model integrated proxy model realizes the second-level mapping from design parameters to pollution response and carbon emission indicators. At the same time, a fine-grained CFD verification mechanism is set up for high uncertainty, critical thresholds and key schemes, which takes into account the efficiency of multi-scheme comparison and the evaluation accuracy of key schemes, and solves the problem that traditional CFD models have high computational overhead and cannot support rapid comparison of a large number of schemes.

[0018] 4. This invention constructs a multi-constraint coupled intelligent screening mechanism for solutions, which incorporates engineering constraints and environmental objectives such as pollution control thresholds, carbon emission thresholds, airspace requirements, land use boundaries, budgets, and construction periods into a unified screening framework. Through hard constraint elimination and multi-objective optimization, it automatically outputs recommended and alternative solutions that meet multi-dimensional requirements, achieving deep coupling between pollution reduction and carbon reduction synergy objectives and engineering implementation boundaries, and solving the problem of the disconnect between environmental objectives and engineering realities in existing technologies.

[0019] 5. This invention quantifies the marginal contribution of each design variable through interpretable artificial intelligence analysis. Combined with multi-dimensional visualization results and a full-process audit traceability mechanism, it enhances the physical interpretability and engineering usability of the solution results, making it easier for planning and design personnel to understand and accept. At the same time, it realizes dynamic re-evaluation and rolling updates of the solution through incremental learning, which can adapt to the dynamic changes in scenario conditions and design requirements, and has strong engineering practicality and scenario adaptability. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments will be briefly introduced below.

[0021] Figure 1 This is a flowchart illustrating the method for selecting pollution and carbon reduction schemes for elevated roads based on UAV three-dimensional monitoring and AI collaborative decision-making, as described in a specific embodiment of the present invention. Detailed Implementation

[0022] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be particularly noted that the following embodiments are for illustrative purposes only and do not limit the scope of the invention. Similarly, the following embodiments are only some, not all, embodiments of the present invention, and all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] See attached document Figure 1 As shown in the figure, this embodiment provides a method for screening pollution reduction and carbon reduction schemes for elevated roads based on UAV three-dimensional monitoring and AI collaborative decision-making, including the following steps: Step S1: Construction of a high-fidelity reference flow field based on "measurement-inversion" two-way coupling S1.1: By using a particulate matter monitoring module, a micro-meteorological monitoring module, a lidar sensor, an image acquisition module, and an airborne IMU / GNSS integrated navigation module mounted on the UAV, multi-height, cross-sectional three-dimensional spatial surveys are conducted on the target elevated road area. Real-time data collection includes particulate matter concentration, wind speed and direction, temperature and humidity, three-dimensional point clouds of the bridge and surrounding buildings, on-site images, and UAV attitude and trajectory data. When necessary, the low-altitude survey data is synchronized and the results are verified by combining auxiliary monitoring equipment deployed on the ground.

[0024] The particulate matter monitoring module is an airborne monitoring device used to measure the mass concentration of suspended particulate matter per unit volume of air, and its output includes at least PM2.5. 1.0 PM 2.5 PM 10 The system includes: particulate matter concentration information; a micro-meteorological monitoring module for synchronously acquiring local meteorological parameters such as wind speed, wind direction, ambient temperature, and relative humidity; a lidar sensor for acquiring spatial distance information of the target area and forming three-dimensional point cloud data using the principle of laser ranging; a set of data consisting of a large number of discrete points with spatial coordinates for characterizing the bridge structure outline, the spatial morphology under the bridge, and the surface features of surrounding buildings; an image acquisition module for acquiring bridge surface information, surrounding building interface information, and on-site spatial morphology image information of the target area; and an airborne IMU / GNSS integrated navigation module for acquiring the attitude information, spatial position, and flight trajectory information of the UAV, including at least pitch angle, roll angle, and heading angle.

[0025] S1.2: Synchronously process multi-source observation data according to a unified time reference. Through coordinate transformation, spatial interpolation, point cloud registration and profile reconstruction, map the data from UAVs, ground sensors and static planning data to the three-dimensional raster coordinate system or parameterized scene unit of the target elevated road area.

[0026] Among them, unified time reference calibration and time series registration are to synchronize the acquisition time, recording sequence and update cycle of observation data from different sources to ensure that various types of data have corresponding relationships under the same time reference.

[0027] Static planning data is used to characterize the target elevated road area, including planning and construction conditions, spatial boundary conditions, and land use attributes. It includes at least one or more of the following: road design drawings, as-built drawings, BIM models, GIS base maps, topographic data, land boundaries, regulatory detailed planning data, building outlines and height data, road red line data, bridge structural dimensions data, municipal ancillary facility layout data, and under-bridge space functional zoning data. Specifically, road design drawings and as-built drawings are used to extract structural parameters such as the elevated road's alignment, bridge deck width, pier locations, clearance height, and ramp layout; BIM models and GIS base maps are used to extract the three-dimensional spatial relationships between the bridge, road, buildings, and surrounding site; land boundaries, regulatory detailed planning data, and road red line data are used to limit the spatial feasibility of candidate schemes; building outlines and height data, under-bridge space functional zoning data, and municipal ancillary facility layout data characterize the surrounding building forms and under-bridge space utilization conditions of the target area; and parametric scene units are sets of structured parameters extracted according to the road, bridge, building, and under-bridge space morphology.

[0028] S1.3: After completing the unified mapping of multi-source data in step S1.2, the basic features of the target area are extracted by combining road design drawings, BIM models, GIS base maps, land boundaries, population activity distribution, sensitive building locations, traffic flow and vehicle emission factor databases, such as elevated height, bridge deck width, pier spacing, under-bridge clearance, under-bridge interface openness, building setback distance, building height gradient, roadside green belt width, roadside functional material configuration and traffic operation status. The mapped pollutant concentration data, micro-meteorological data and three-dimensional spatial morphology data are then rasterized, spatially interpolated, attribute-associated and standardized for storage to obtain the pollutant concentration, meteorological parameters, spatial morphology parameters, traffic operation parameters and emission parameters corresponding to each three-dimensional raster unit, thereby forming a high-resolution pollution field database.

[0029] The high-resolution pollution field database uses three-dimensional raster cells or parametric scene cells as basic storage objects. It integrates a multi-source standardized data set containing information such as pollutant concentration, local meteorology, road and bridge structure, surrounding building morphology, traffic operation status, and emission intensity. This data is used for subsequent scheme generation, rapid evaluation, and dynamic updates. Specifically, GIS software can be used to complete coordinate unification and raster division, point cloud processing software can be used to complete point cloud registration and scene reconstruction, and Python, Matlab, or other data processing programs can be used to implement spatial interpolation, raster assignment, and pollution field reconstruction. The preferred spatial interpolation method is Kriging interpolation or inverse distance weighting.

[0030] S1.4: Using high-resolution pollution field data obtained from UAV field measurements as the calibration benchmark, the CFD model of the monitored elevated road scenario is adaptively calibrated to improve the accuracy of the simulation results in representing the actual pollution diffusion characteristics. After completing the model calibration, the Latin hypercube sampling method is used to perform multiple sets of combined sampling of elevated road geometric parameters, traffic operation parameters, wind field parameters, and surrounding building parameters. Based on the calibrated CFD model, corresponding pollution diffusion results and carbon emission results are generated, thus forming a high-fidelity simulation sample library.

[0031] After forming a high-fidelity simulation sample library, the elevated geometric parameters, traffic operation parameters, wind field parameters, surrounding building parameters, and ancillary facility parameters of each group of samples are used as input features. The corresponding pollution diffusion indicators, carbon emission indicators, ventilation performance indicators, or concentration distribution results of virtual monitoring points in the target area are used as output labels to construct a training sample set for the surrogate model. Subsequently, the training sample set is normalized and divided into training subsets and validation subsets. Support vector regression models, XGBoost models, and deep neural networks with attention mechanisms are trained on these subsets to obtain multiple base learners. Then, the prediction results of each base learner are used as secondary inputs, and a meta-learner (preferably a ridge regression model) is used to perform weighted fusion to obtain a multi-model integrated surrogate model for rapid evaluation of candidate solutions. During the training process, cross-validation is preferred to optimize the model hyperparameters to improve the model's generalization ability in different scenarios.

[0032] The above steps generate a simulated sample library and corresponding surrogate models that have been measured and calibrated, which can be used for rapid evaluation and screening of subsequent candidate solutions.

[0033] Step S2: Structured Construction of Design Objectives and Screening Criteria After forming the measured and calibrated simulation sample library and corresponding proxy model in step S1, based on the pre-built domain knowledge base, regulatory clause library, engineering capability library, carbon emission factor library, and historical case library, the user-input design objectives, constraints, and optimization preferences are understood, parsed, and structured to form the target constraint set required for subsequent candidate solution generation, rapid evaluation, and multi-objective optimization; specifically, the following steps are included: S2.1: Receive design objectives, constraints, and preference inputs from planners, design units, construction units, or management departments. Preference inputs may be provided in natural language text, structured forms, or preset options. The design objectives should include at least one or more of the following: target area, design stage, pollution control objectives, carbon emission control objectives, cost control objectives, under-bridge space utilization objectives, sensitive target protection requirements, and construction implementation objectives; the constraints should include at least one or more of the following: airspace requirements, land boundaries, budget limits, construction period requirements, structural modification restrictions, traffic organization restrictions, and operation and maintenance conditions; the target area should preferably correspond to one of the following typical scenarios: high-rise buildings on both sides, high-rise buildings on one side and open space on the other side, and open space on both sides.

[0034] Users can directly select the typical scene of the target area, or input the height of surrounding buildings, building setback distance, bridge structure dimensions and traffic operation parameters, and the system will automatically complete the scene recognition.

[0035] S2.2: The input design goals, constraints, and preferences are structurally extracted, unit normalized, nouns disambiguated, and consistency checked, and then transformed into calculable target parameters, constraint parameters, evaluation indicators, and design variable boundaries. Among them, structural extraction is to transform text descriptions, form inputs, or preset rules into a set of parameters that can be directly called upon in subsequent candidate solution generation, evaluation, and screening. This includes semantic parsing of user input, parameter slot extraction, and parameter mapping to identify key parameters such as target areas, typical scene types, building height, building setback distance, bridge structure dimensions, pollution control thresholds, carbon emission thresholds, and engineering constraints.

[0036] The design variable boundary represents the adjustable range, prohibition conditions, and value restrictions set for the main form of the elevated road, the treatment of the space under the bridge, the configuration of ancillary facilities, the control of surrounding buildings, the configuration of greening, and traffic organization parameters.

[0037] S2.3: Based on the pre-built domain knowledge base, regulatory clause base, engineering capability base, carbon emission factor base and historical case base, and combined with the preset rule templates and constraint matching rules, the target parameters, constraint parameters, evaluation indicators and design variable boundaries are completed, verified and organized; The process involves identifying and extracting hard constraints such as airspace requirements, land use boundaries, safety distances, and environmental quality targets based on a regulatory clause database; identifying the applicable scope and prohibitions for different construction methods, structural modification methods, material configuration methods, and ancillary facility layout methods based on an engineering capability database; supplementing carbon emission accounting parameters corresponding to various construction behaviors and material schemes based on a carbon emission factor database; and extracting high-performance scheme characteristics and optimization experiences similar to the current scenario based on a historical case database. Finally, based on the user's input optimization preferences, the pollution control target, carbon emission target, cost target, construction period target, and space utilization target are weighted and allocated to form a set of scheme selection conditions that match the scenario characteristics, design objectives, and constraints of the target area.

[0038] The engineering capability library is a rule base formed by structuring and organizing the applicable conditions, adjustable ranges, and insurmountable engineering boundaries of different construction methods, materials, and ancillary facilities. The scheme screening condition set includes at least one or more of the following: the range of selectable design variables, hard constraints, soft target weights, evaluation index set, sensitive target protection rules, and prohibitive conditions. The weight allocation method is preferably one or more of the following: the analytic hierarchy process, the entropy weight method, or preset weight mapping rules. The organization and sorting of screening conditions preferably adopt the weighted scoring method, the TOPSIS method, or the rule-based sorting method.

[0039] In this embodiment, the process of completing, verifying, and organizing the target parameters, constraint parameters, evaluation indicators, and design variable boundaries is preferably implemented collaboratively by the natural language processing module, rule engine, and database management system. More preferably, it is implemented by a large language model combined with a domain knowledge base, a legal clause base, and an engineering capability base for semantic parsing and rule verification.

[0040] S2.4: When there are conflicts between input objectives, key constraints are missing, or variable boundaries are unclear, generate items to be confirmed and prompt for manual supplementation or revision. For example, when a user simultaneously requests that the underpass clearance remain unchanged, the building setback not increase, and the pedestrian level PM... 2.5 When the increment decreases significantly, potential conflicts are identified and adjustable variables and their relaxation suggestions are given. After manual confirmation, the target parameters, constraint parameters and scheme screening condition set are updated again until a structured task input that can be used for subsequent candidate scheme generation, pollution reduction and carbon reduction assessment and multi-objective selection is formed.

[0041] The above steps generate a structured set of objectives and constraints for candidate scheme generation, pollution reduction and carbon reduction assessment, and multi-objective screening.

[0042] Step S3: Multiple verifications and supplementary confirmations of structured target constraints S3.1: Input the structured target constraint set formed in step S2 into the regulatory compliance verification submodule, the engineering feasibility verification submodule, the spatial constraint verification submodule, and the data applicability verification submodule in sequence, and perform rule matching and constraint detection on the target parameters, constraint parameters, evaluation indicators, and design variable boundaries.

[0043] Among them, the regulatory compliance verification submodule is used to verify the legality of road design specifications, clearance requirements, land use boundaries, safety distances, environmental quality targets, and requirements for the use of space under bridges; the data applicability verification submodule is used to mark design requirements that rely on low-reliability data or exceed the applicable conditions of the model, based on monitoring density, scenario completeness, and model applicability.

[0044] S3.2: When constraint conflicts, missing parameters, unclear boundaries, or exceeding engineering capabilities are detected, feedback information is generated, including the source of the conflict, the violated conditions, the affected variables, and alternative suggestions, prompting manual confirmation. The feedback information should at least include unsatisfactory hard constraints, relaxable variables, suggested adjustment directions, and their potential environmental and engineering impacts.

[0045] S3.3: Re-execute structured updates and constraint checks on the revised objectives and constraints that have been manually confirmed until the structured objective and constraint set passes all checks; if there are irremovable conflicts, retain the conflicting items and limit the scope of variables in the subsequent candidate solution generation process.

[0046] S3.4: Organize the verified target parameters, constraint parameters, evaluation indicators, and design variable boundaries into inputs for the scheme generation and screening task; the inputs for the scheme generation and screening task shall include at least the scenario call conditions corresponding to the target area, the set of adjustable design variables, the range of variable values, the basic prohibition conditions, the set of evaluation indicators, and the target weight information required for subsequent optimization; the target weight information is the relative priority or weight coefficient assigned to pollution control targets, carbon emission control targets, cost control targets, construction period targets, under-bridge space utilization targets, and sensitive target protection targets.

[0047] The scene invocation conditions include at least the scene type of the target area, the basic geometric information of the bridge and road, the layout characteristics of surrounding buildings, the applicable local meteorological conditions, the traffic operation status, and the distribution information of sensitive targets; the set of adjustable design variables includes at least the main morphology variables of the elevated road, the processing variables of the space under the bridge, the configuration variables of the ancillary barriers, the control variables of the surrounding buildings, the configuration variables of greening or functional materials, and the traffic organization coordination variables; the variable value range is the upper limit, lower limit, discrete optional value, or continuous variation range corresponding to each adjustable design variable; the basic prohibition conditions include at least the conditions of violating the clearance requirements, exceeding the land use boundary, exceeding the budget limit, exceeding the construction period limit, violating the structural safety requirements, violating the traffic organization requirements, or triggering the protection restrictions of sensitive targets; the set of evaluation indicators includes at least one or more of the following: pollutant concentration indicators, pollution exposure indicators, ventilation performance indicators, carbon emission indicators, engineering implementation indicators, and space utilization indicators; The above task inputs can be automatically generated by the rule engine, database management system and data processing program, and can be directly called by the subsequent candidate solution generation module and multi-objective optimization module.

[0048] The above steps generate a structured set of objective constraints that has passed multiple constraint checks, as well as input for the scheme generation task.

[0049] Step S4: Generation of candidate elevated road construction or renovation plans S4.1: Based on the verified structured objective constraint set, determine the design variable boundaries, variable value ranges, and prohibitive conditions. Combined with the site conditions extracted from the high-resolution contaminated field database formed in step S1 and the task inputs formed in steps S2 and S3, construct a search space for candidate elevated road construction or renovation schemes. The search space should include at least the main form of the elevated road, the openness of the space under the bridge, the configuration of ancillary barriers, the control methods for surrounding buildings, greening or functional material strategies, and traffic organization coordination strategies.

[0050] S4.2: Based on the input of the scheme generation and screening task formed in step S3 and the candidate scheme search space constructed in step S4.1, firstly, read the set of adjustable design variables, variable value range, basic prohibition conditions, and target weight information; then, parameterize and encode each design variable according to the preset combination rules, and perform a combination search on the adjustable design variables under the premise of meeting the requirements of airspace, land boundary, budget constraints, structural safety requirements, and sensitive target protection requirements to generate an initial candidate scheme set; then, call the historical case similarity retrieval module to identify existing cases with similar scene characteristics to the current target area, extract the variable combination characteristics of high-performance schemes, and prioritize or reduce the search space accordingly; finally, combine heuristic optimization strategies or preference-guided strategies to iteratively expand, screen, and sort the initial candidate scheme set to obtain multiple candidate schemes that meet the constraints.

[0051] For the scenario of newly constructed elevated roads, candidate solutions may include different elevated heights, bridge deck widths, pier forms, building setback distances, and ventilation organization combinations; for the scenario of existing elevated road renovation, candidate solutions may include different under-bridge open designs, barrier permeability, local wind-guiding components, green barrier configurations, and traffic organization adjustment strategies; the above candidate solutions are preferably generated under the spatial morphological conditions, traffic operation conditions, and engineering constraints of the typical scenario in the target area.

[0052] S4.3: Standardize the coding of candidate schemes, and represent each scheme as a parameter vector or graph structure consisting of geometric variables, environmental variables, material variables, traffic variables and control variables. Eliminate significantly infeasible schemes based on basic constraints such as airspace, construction, land boundary, budget and operation and maintenance conditions.

[0053] S4.4: Establish a binding relationship between candidate solutions and target area scenarios, so that the scenario database, simulation sample library and proxy model can be called according to the "solution-scenario" correspondence during the subsequent evaluation process, and the comparability evaluation of different candidate solutions under the same scenario conditions can be achieved.

[0054] The above steps generate a set of preliminary screening candidate elevated road construction or renovation schemes.

[0055] Step S5: Rapid Assessment of the Synergistic Effect of Pollution Reduction and Carbon Reduction S5.1: Match and merge the parameters of the preliminary screening candidate schemes with the background parameters of the typical scenarios in the target area to establish a collaborative assessment input for pollution reduction and carbon reduction for each candidate scheme. The background parameters are composed of surrounding building heights, building setback distances, bridge structure dimensions, traffic operation parameters, local meteorological parameters, background pollution parameters, and spatial morphology parameters corresponding to the identified typical scenarios, either input by the user or extracted from planning data. Matching and merging refers to corresponding, supplementing, integrating, and standardizing the candidate scheme parameters and scenario background parameters according to a unified field structure to form a standardized assessment input for each candidate scheme.

[0056] The collaborative assessment inputs include at least one or more of the following: the main morphological parameters of the elevated road, the parameters for handling the space under the bridge, the morphological parameters of the surrounding buildings, the traffic operation parameters, the local meteorological parameters, the background pollution parameters, and the carbon emission-related parameters.

[0057] S5.2: Based on the high-fidelity simulation sample library, reduced-order physical model or data-driven equivalent operator formed in step S1.4, evaluate the pollution transport process and ventilation performance of candidate schemes in the target area; wherein, preferably, the surrogate model trained in step S1.4 is used to quickly extrapolate multiple candidate schemes and calculate the pollutant concentration distribution, exposure risk index, under-bridge space exchange capacity, local retention intensity and pollution plume diffusion capacity corresponding to the candidate schemes.

[0058] When the relative uncertainty of the prediction result of a candidate scheme by the proxy model is greater than 10%, or the absolute value of the deviation between the corresponding index value of the candidate scheme and the user-set screening threshold is less than 5% of the screening threshold, or a design variable of the candidate scheme exceeds 10% of the upper and lower bounds of the corresponding variable value in the high-fidelity simulation sample library, or the candidate scheme is ranked in the top 3 in the comprehensive ranking, the CFD mechanism model calibrated by UAV in step S1.4 is called to conduct a fine review and evaluation of the candidate scheme; the review and evaluation includes importing the collaborative evaluation input into the CFD mechanism model and recalculating the local wind field characteristics, pollutant concentration field, pedestrian layer concentration, building window height concentration, excess volume ratio in the area under the bridge, and ventilation performance index corresponding to the candidate scheme.

[0059] The CFD mechanism model is preferably implemented using ANSYS Fluent, OpenFOAM or other computational fluid dynamics software; the continuity equation, momentum equation and pollutant convection and diffusion equation can be solved using the finite volume method; the turbulence model is preferably the standard k-ε model.

[0060] Specifically, the review and evaluation method is as follows: the collaborative evaluation input is imported into the CFD mechanism model to solve for the local wind field and pollutant concentration field corresponding to the candidate scheme. By slicing, extracting, statistically analyzing and calculating the indicators of the flow field and concentration field, the concentration at the pedestrian level, the concentration at the building window height, the proportion of the volume exceeding the standard in the area under the bridge and the ventilation performance indicators are obtained.

[0061] S5.3: A coordinated quantification of carbon emissions during the construction and operation phases of the candidate projects shall be conducted. Construction phase carbon emissions shall include at least the carbon emissions from bridge materials, ancillary components, landscaping facilities, and functional materials. Operation phase carbon emissions shall include at least the carbon emissions generated from changes in traffic flow, energy consumption of ancillary facilities, and maintenance activities. This coordinated quantification integrates pollution control benefits with carbon emission costs into a unified assessment framework for joint comparison.

[0062] S5.4: Use a pre-trained surrogate model to quickly extrapolate multiple candidate schemes, output pollution diffusion indicators, carbon emission indicators, exposure risk indicators, ventilation performance indicators, cost indicators and engineering implementation indicators corresponding to each candidate scheme, and simultaneously provide the model prediction confidence or uncertainty evaluation; wherein, the surrogate model is the multi-model integrated surrogate model obtained by training based on the high-fidelity simulation sample library in step S1.4.

[0063] S5.5: Based on interpretable artificial intelligence analysis, combined with SHAP value analysis, attention weight analysis, sensitivity analysis or counterfactual interpretation, quantify the degree of influence of each design variable on pollution control effect and carbon emission results, so as to identify the marginal contribution of key variables such as elevated height, building setback distance, under-bridge openness, barrier permeability, greening configuration and traffic organization parameters, and form a multi-index evaluation result matrix of candidate schemes and parameter contribution information.

[0064] In this embodiment, the multi-index evaluation result matrix and parameter contribution information are generated by uniformly numbering, mapping, and summarizing the evaluation results and interpretation results of each candidate scheme, and are used for threshold screening and multi-objective optimization in the subsequent step S6.

[0065] The multi-index evaluation result matrix is ​​a structured set of results consisting of candidate schemes as rows and pollution diffusion indicators, exposure risk indicators, ventilation performance indicators, carbon emission indicators, cost indicators, and engineering implementation indicators as columns. The index values ​​in each column are obtained by extracting, calculating, and standardizing the corresponding model evaluation results in steps S5.2 to S5.4.

[0066] The parameter contribution information is a quantitative result based on the SHAP value analysis, attention weight analysis, sensitivity analysis or counterfactual interpretation results in step S5.5, which is the degree of influence of design variables such as elevated height, building setback distance, under-bridge openness, barrier permeability, greening configuration and traffic organization parameters.

[0067] Step S6: Scheme selection and multi-objective optimization based on set indicator thresholds S6.1: Receive the target indicator threshold and engineering constraints set by the user, and match them with the multi-indicator evaluation result matrix output in step S5.

[0068] Among them, the target indicator thresholds include at least one or more of the following: pedestrian level pollutant concentration threshold, building window height pollution threshold, bridge underpass area exceeding the standard volume ratio threshold, carbon emission per unit length threshold, unit pollution reduction benefit threshold, and budget ceiling; the engineering constraints include at least one or more of the following: airspace constraints, land use boundary constraints, structural modification restrictions, construction period constraints, and construction accessibility constraints.

[0069] S6.2: Perform hard constraint screening on candidate solutions. When a candidate solution fails to meet the requirements of any constraint such as pollution threshold, carbon emission threshold, airspace, red line, cost, construction period, or construction accessibility, the candidate solution is removed from the set of feasible solutions to obtain a set of feasible solutions that meet the basic requirements.

[0070] S6.3: Perform multi-objective optimization on the set of feasible solutions that meet the hard constraints; wherein, multi-objective optimization can be accomplished by weight-based comprehensive scoring ranking, Pareto front non-dominated ranking, rule priority recommendation or a combination thereof; when users emphasize pollution reduction effect, increase the target weight of pollution control related indicators; when users emphasize carbon reduction and cost, increase the target weight of carbon emission indicators and cost indicators.

[0071] S6.4: Based on the multi-objective optimization results, design strategies are extracted from the set of feasible solutions. These design strategies include not only the final recommended scheme, but also corresponding suggestions on the elevated main structure form, the treatment of the space under the bridge, the control of surrounding buildings, the configuration of green spaces and functional materials, and the coordination of traffic organization.

[0072] The above steps generate recommended methods for the construction or renovation of elevated roads, a sequence of alternative solutions, and design strategies.

[0073] Step S7, Result Visualization, Full-Process Audit and Dynamic Updates S7.1: Generate visualization results for the recommended and alternative schemes output in step S6 to support planners, designers, or managers in understanding the results from multiple dimensions, including spatial distribution, exposure hotspots, carbon emission composition, and comparison of scheme advantages and disadvantages. The visualization results should include at least one or more of the following: pollution diffusion cloud map, exposure hotspot map, carbon emission comparison map, scheme radar chart, and parameter contribution interpretation chart.

[0074] S7.2: Unify the identification and storage of the target area scenario version, structured target constraint set version, constraint verification record, candidate solution version, evaluation model version, screening results and final recommended solution to form a full-process audit chain; where the full-process audit chain refers to the set of associated records that can trace back the source of requirements, the source of scenario data, the configuration of key parameters, the model call relationship and the output results.

[0075] S7.3: When the target area monitoring data, traffic conditions, weather conditions, scene boundary conditions or user-defined thresholds change, the system automatically calls the updated scene database and performs incremental learning or migration updates on the proxy model based on the newly added monitoring samples. It then re-executes the candidate solution evaluation, feasibility screening and multi-objective optimization steps to achieve dynamic re-evaluation and rolling updates of the solution.

[0076] In summary, this invention constructs a pollution reduction and carbon reduction scheme selection method for three typical elevated road scenarios: two sides with high-rise buildings, one side with high-rise buildings and one side with open space, and two sides with open space. This method employs a "UAV three-dimensional perception - intelligent semantic analysis - intelligent generation of candidate schemes - measured calibration and simulation - interpretable AI rapid evaluation - target constraint screening and optimization - dynamic learning and updating" approach. This method goes beyond simply analyzing the pollution diffusion mechanism of elevated road traffic; it addresses engineering design and planning decision-making scenarios by organically coupling high spatiotemporal resolution three-dimensional pollution field data acquired by UAVs, a measured and calibrated mechanism model, a surrogate model based on multi-model integration, and a multi-objective constraint screening mechanism. This enables the automatic generation, rapid evaluation, feasibility screening, and strategy output of schemes for the construction, reconstruction, expansion, and underpass space optimization of elevated roads, given pollution control indicators, carbon emission indicators, and engineering boundary conditions.

[0077] Furthermore, the functional units in the embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0078] The above description is only a part of the embodiments of the present invention and does not limit the scope of protection of the present invention. Any equivalent device or equivalent process transformation made based on the content of the present invention specification and drawings, or direct or indirect application in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for screening pollution reduction and carbon reduction schemes for elevated roads based on UAV three-dimensional monitoring and AI collaborative decision-making, characterized in that, Includes the following steps: S1: The target elevated road area is surveyed in three dimensions using multiple monitoring modules carried by the UAV to obtain multi-source observation data, which is then integrated to construct a high-resolution pollution field database. The CFD model is calibrated based on the measured data to generate a high-fidelity simulation sample library and train a multi-model integrated surrogate model. S2: Receive user input of design goals, constraints, and preference preferences, perform structured analysis and transformation, and combine them with a pre-built domain knowledge base, regulatory clause base, engineering capability base, carbon emission factor base, and historical case base to form a structured set of goals and constraints required for candidate solution generation and screening; S3: The structured target constraint set is sequentially checked through the multi-dimensional verification submodule. Parameters with conflicts or missing parameters are corrected and feedback is provided until a solution that has passed all verifications is generated and the task input for filtering is formed. S4: Based on the task input verified, construct a candidate solution search space, perform combined search and optimization on adjustable design variables, and generate and pre-screen a set of candidate solutions that meet the basic constraints. S5: Use the proxy model obtained in step S1 to evaluate the candidate solutions, call the CFD model to verify the solutions that meet the preset conditions, and combine interpretable artificial intelligence to quantify the contribution of each design variable to form a multi-index evaluation result matrix. S6: Based on preset index thresholds and engineering constraints, perform hard constraint screening and multi-objective optimization on candidate solutions, and output recommended solutions, alternative sequences and supporting design strategies. S7: Generate visualized solution results and establish a traceable audit chain throughout the entire process to enable dynamic re-evaluation and rolling updates of the solution.

2. The method for screening pollution reduction and carbon reduction schemes for elevated roads based on UAV three-dimensional monitoring and AI collaborative decision-making as described in claim 1, characterized in that, Step S1 specifically includes the following steps: S1.1: Use UAVs equipped with multiple monitoring modules to conduct multi-height, cross-sectional three-dimensional spatial surveys of the target elevated road area, collect particulate matter concentration, local meteorological parameters, three-dimensional point cloud, on-site images and UAV attitude trajectory data, and verify the data in conjunction with ground auxiliary monitoring equipment when necessary; S1.2: Synchronously process multi-source observation data according to a unified time reference. Through coordinate transformation, spatial interpolation, point cloud registration and profile reconstruction, map the data from UAVs, ground sensors and static planning data to the three-dimensional raster coordinate system or parameterized scene unit of the target elevated road area. S1.3: Combine the road design drawings, BIM model, GIS base map, land boundary, traffic flow and vehicle emission factor database of the target area to extract the basic features of the elevated roads and surrounding environment of the target area; The mapped pollutant concentration data, micrometeorological data, and three-dimensional spatial morphology data are rasterized, spatially interpolated, attribute-associated, and standardized for storage to obtain multi-dimensional parameters corresponding to each three-dimensional raster unit, thereby forming a high-resolution pollution field database. S1.4: Using high-resolution pollution field data obtained from UAV field measurements as the calibration benchmark, the CFD model of the target elevated road scene is adaptively calibrated; multiple sets of combined sampling are performed on the elevated road geometric parameters, traffic operation parameters, wind field parameters and surrounding building parameters through the Latin hypercube sampling method; based on the calibrated CFD model, the corresponding pollution diffusion results and carbon emission results are generated, and a high-fidelity simulation sample library is constructed. S1.5: The elevated geometric parameters, traffic operation parameters, wind field parameters, surrounding building parameters, and ancillary facility parameters of each group of samples are used as input features, and the corresponding pollution diffusion indicators, carbon emission indicators, ventilation performance indicators, or concentration distribution results of virtual monitoring points in the target area are used as output labels to construct a training sample set for the surrogate model; then the training sample set is normalized and divided into a training subset and a validation subset, and a support vector regression model, an XGBoost model, and a deep neural network with attention mechanism are trained respectively to obtain multiple base learners; Then, the prediction results of each base learner are used as secondary inputs, and a meta-learner is used to weight and fuse them to obtain a multi-model ensemble proxy model for rapid evaluation of candidate solutions.

3. The method for screening pollution reduction and carbon reduction schemes for elevated roads based on UAV three-dimensional monitoring and AI collaborative decision-making as described in claim 2, is characterized in that... The static planning data mentioned in step S1.2 shall include at least one or more of the following: road design drawings, as-built drawings, BIM model, GIS base map, topographic data, land boundary, regulatory detailed planning data, building outline and height data, road red line data, bridge structure dimension data, municipal ancillary facility layout data, and functional zoning data of the space under the bridge. The parameterized scene unit is a set of structured parameters extracted according to the spatial morphology of roads, bridges, buildings, and the space under bridges.

4. The method for screening pollution reduction and carbon reduction schemes for elevated roads based on UAV three-dimensional monitoring and AI collaborative decision-making as described in claim 1, characterized in that, Step S2 specifically includes the following steps: S2.1: Receive user input regarding design goals, constraints, and preference criteria; S2.2: Perform structured extraction, unit normalization, noun disambiguation, and consistency verification on the input design objectives, constraints, and optimization preferences, and transform them into calculable target parameters, constraint parameters, evaluation indicators, and design variable boundaries; S2.3: Based on the pre-built domain knowledge base, regulatory clause base, engineering capability base, carbon emission factor base and historical case base, and combined with the preset rule templates and constraint matching rules, the target parameters, constraint parameters, evaluation indicators and design variable boundaries are completed, verified and organized, and the weight allocation of multiple objectives is completed according to the user's preference to form a set of scheme selection conditions; S2.4: When there are conflicts among input objectives, key constraints are missing, or variable boundaries are unclear, generate items to be confirmed and prompt manual supplementation or revision until a structured task input that can be used for subsequent candidate solution generation, pollution reduction and carbon reduction assessment, and multi-objective selection is formed.

5. The method for screening pollution reduction and carbon reduction schemes for elevated roads based on UAV three-dimensional monitoring and AI collaborative decision-making as described in claim 4, characterized in that, The design objectives mentioned in step S2.1 include at least one or more of the following: target area, design stage, pollution control objectives, carbon emission control objectives, cost control objectives, under-bridge space utilization objectives, sensitive target protection requirements, and construction implementation objectives; the constraints include at least one or more of the following: airspace requirements, land use boundaries, budget limits, construction period requirements, structural modification restrictions, traffic organization restrictions, and operation and maintenance conditions. In step S2.2, structured extraction involves semantic parsing, parameter slot extraction, and parameter mapping of user input to identify key parameters including target area, typical scene type, building height, building setback distance, bridge structure dimensions, pollution control threshold, carbon emission threshold, and engineering constraints. The design variable boundary refers to the adjustable range, prohibition conditions, and value restrictions set for the main form of the elevated road, the treatment of the space under the bridge, the configuration of ancillary facilities, the control method of surrounding buildings, the greening configuration method, and traffic organization parameters.

6. The method for screening pollution reduction and carbon reduction schemes for elevated roads based on UAV three-dimensional monitoring and AI collaborative decision-making as described in claim 1, characterized in that, Step S3 specifically includes the following steps: S3.1 Input the structured target constraint set formed in step S2 into the regulatory compliance verification submodule, the engineering feasibility verification submodule, the spatial constraint verification submodule, and the data applicability verification submodule in sequence, and perform rule matching and constraint detection on the target parameters, constraint parameters, evaluation indicators, and design variable boundaries; The aforementioned regulatory compliance verification involves legally verifying road design specifications, clearance requirements, land use boundaries, safety distances, environmental quality targets, and requirements for the use of space under bridges. The data applicability verification involves combining monitoring density, scenario completeness, and model applicability range to mark designs that rely on low-reliability data or exceed the model's applicable conditions. S3.2: When a constraint conflict, missing parameter, unclear boundary, or exceeding the engineering capability is detected, the system generates feedback information including the source of the conflict, the violation condition, the affected variables, and alternative suggestions, and prompts for manual confirmation. S3.3: Re-execute structured updates and constraint checks on the revised objectives and constraints that have been manually confirmed until the structured objective and constraint set passes all checks; if there are irremovable conflicts, retain the conflicting items and limit the scope of variables in the subsequent candidate solution generation process; S3.4: Organize all verified target parameters, constraint parameters, evaluation indicators, and design variable boundaries into inputs for scheme generation and screening tasks.

7. The method for screening pollution reduction and carbon reduction schemes for elevated roads based on UAV three-dimensional monitoring and AI collaborative decision-making as described in claim 1, characterized in that, Step S4 specifically includes the following steps: S4.1: Based on the verified structured objective constraint set, determine the design variable boundaries, variable value ranges, and prohibitive conditions. Combined with the site conditions extracted from the high-resolution contaminated field database formed in step S1 and the task inputs formed in steps S2 and S3, construct a search space for candidate elevated road construction or renovation schemes. The search space includes at least the main form of the elevated road, the opening method of the space under the bridge, the configuration of the auxiliary barriers, the control method of surrounding buildings, the greening or functional material strategy, and the traffic organization coordination strategy. S4.2: First, read the set of adjustable design variables, the range of variable values, the basic prohibition conditions, and the target weight information; then, according to the preset combination rules, parameterize and encode each design variable, and under the premise of meeting the requirements of clearance, land boundary, budget constraints, structural safety requirements, and sensitive target protection requirements, perform a combination search on the adjustable design variables to generate an initial candidate scheme set; Subsequently, the historical case similarity retrieval module is invoked to extract the variable combination features of high-performing solutions in similar scenarios, and the search space is preferentially sampled or reduced accordingly. Finally, combined with heuristic optimization strategies or preference-guided strategies, the initial candidate solution set is iteratively expanded, filtered, and sorted to obtain multiple candidate solutions that meet the constraints. S4.3: Standardize and code the candidate solutions, representing each solution as a parameter vector or graph structure consisting of geometric variables, environmental variables, material variables, traffic variables, and control variables, and eliminate infeasible solutions based on basic constraints; among which, basic constraints include clearance, construction, land boundary, budget, and operation and maintenance conditions; S4.4: Establish the binding relationship between candidate solutions and target area scenarios to form a set of candidate solutions that have undergone preliminary screening.

8. The method for screening pollution reduction and carbon reduction schemes for elevated roads based on UAV three-dimensional monitoring and AI collaborative decision-making as described in claim 1, characterized in that, Step S5 specifically includes the following steps: S5.1: Match and merge the parameters of the preliminary screening candidate schemes with the background parameters of typical scenarios in the target area to establish a collaborative assessment input for pollution reduction and carbon reduction for each candidate scheme; The background parameters are composed of surrounding building heights, building setback distances, bridge structure dimensions, traffic operation parameters, local meteorological parameters, background pollution parameters, and spatial morphology parameters corresponding to the identified typical scenarios, either input by the user or extracted from planning data. The collaborative evaluation input includes at least one or more of the following: elevated road main morphology parameters, bridge under-bridge space processing parameters, surrounding building morphology parameters, traffic operation parameters, local meteorological parameters, background pollution parameters, and carbon emission-related parameters. S5.2: Based on the multi-model integrated proxy model obtained in step S1, the pollution transport process and ventilation performance of the candidate schemes are simulated to obtain the pollutant concentration distribution, exposure risk indicators, and ventilation performance related indicators; S5.3: Conduct coordinated quantification of carbon emissions during the construction and operation phases of candidate projects; wherein, carbon emissions during the construction phase shall include at least the carbon emissions of bridge materials, ancillary components, greening facilities and functional materials, and carbon emissions during the operation phase shall include at least the carbon emissions generated by changes in traffic operation, energy consumption of ancillary facilities operation and maintenance activities; S5.4: Use a pre-trained surrogate model to quickly extrapolate multiple candidate schemes, output pollution diffusion indicators, carbon emission indicators, exposure risk indicators, ventilation performance indicators, cost indicators and engineering implementation indicators corresponding to each candidate scheme, and simultaneously provide the model prediction confidence or uncertainty evaluation. S5.5: Based on interpretable artificial intelligence analysis, combined with SHAP value analysis, attention weight analysis, sensitivity analysis or counterfactual interpretation, quantify the degree of influence of each design variable on pollution control effect and carbon emission results, obtain parameter contribution information, and form a multi-index evaluation result matrix.

9. The method for screening pollution reduction and carbon reduction schemes for elevated roads based on UAV three-dimensional monitoring and AI collaborative decision-making as described in claim 1, characterized in that, Step S6 specifically includes the following steps: S6.1: Receive the target indicator threshold and engineering constraints set by the user, and match them with the multi-indicator evaluation result matrix output in step S5; The target indicator thresholds include at least one or more of the following: pedestrian level pollutant concentration threshold, building window height pollution threshold, bridge underpass area exceeding the standard volume ratio threshold, carbon emission per unit length threshold, unit pollution reduction benefit threshold, and budget ceiling; the engineering constraints include at least one or more of the following: airspace constraints, land use boundary constraints, structural modification restrictions, construction period constraints, and construction accessibility constraints. S6.2: Perform hard constraint screening on candidate solutions, eliminate candidate solutions that do not meet the requirements on any constraint, and obtain a set of feasible solutions that meet the basic requirements; S6.3: Perform multi-objective optimization on the set of feasible solutions, wherein the multi-objective optimization is accomplished by one or more combinations of weight-based comprehensive score ranking, Pareto front non-dominated ranking, and rule priority recommendation; S6.4: Based on the results of multi-objective optimization, extract design strategies from the set of feasible solutions.

10. The method for screening pollution reduction and carbon reduction schemes for elevated roads based on UAV three-dimensional monitoring and AI collaborative decision-making as described in claim 1, characterized in that, The visualization results mentioned in step S7 include at least one or more of the following: pollution diffusion cloud map, exposure heat map, carbon emission comparison map, scheme radar map, and parameter contribution interpretation map; the full-process audit chain is a set of associated records that can trace back the source of demand, the source of scenario data, the configuration of key parameters, the model call relationship, and the output results. The dynamic re-evaluation and rolling update of the scheme are carried out in the following way: when the monitoring data of the target area, traffic conditions, meteorological conditions, scene boundary conditions or user-defined thresholds change, the updated scene database is called, and incremental learning or transfer updates are performed on the proxy model based on the newly added monitoring samples. The candidate scheme evaluation, feasibility screening and multi-objective optimization steps are re-executed to realize the dynamic re-evaluation and rolling update of the scheme.

Images