A method and system for constructing a dynamic soundscape map of urban space and a medium

Abstract

The application provides a kind of urban space dynamic sound scene map construction method, system and medium, belong to the field of dynamic prediction, data integration, space mapping and visual display of urban sound scene, the method comprises: first, by fusing subjective perception score of residents and acoustic, ecological and visual objective environmental data, construct comprehensive sound scene perception index;Further, combined with acoustic simulation and measured correction technology, low-cost generation covers the high-density sound scene training sample set of global;Subsequently, the complex relationship between static space features and dynamic environmental features and sound scene perception index is learned using a machine learning model to establish a dynamic prediction model;Finally, the sound scene perception index prediction value output by the model is rendered through a GIS platform to generate a dynamic sound scene map that reflects the rhythm of urban operation. The present application overcomes the defects of existing sound scene technology such as static, single-dimensional and low resolution, and realizes dynamic, fine and interpretable quantification and visualization of human sound scene experience.

Description

Technical Field

[0001] This invention belongs to the field of dynamic prediction, data integration, spatial mapping and visualization of urban soundscapes, and specifically relates to a method, system and medium for constructing a dynamic soundscape map of urban space. Background Technology

[0002] Soundscape, as a composite concept integrating hearing and environment, centers on people's overall perception and understanding of sound in a specific context. It transcends the traditional focus on physical sound pressure levels in noise control, emphasizing instead the ecological value, psychological impact, and interactive relationship between sound and space and culture. Against this backdrop, dynamic soundscape mapping has emerged, aiming to upgrade the urban sound environment from a static noise distribution background map into a dynamic soundscape health monitoring instrument that reflects the city's operational rhythms in real time. This tool not only serves for the precise location of noise problems but also dedicates itself to diagnosing causes, predicting changes, and simulating intervention effects, thereby providing core decision support for the entire chain of refined urban management and human-centered design, from macro-planning to micro-governance.

[0003] Existing methods for constructing such maps primarily employ machine learning models (such as random forests) to learn the relationship between soundscape and spatial features. However, their model performance is limited by the quality of the training dataset, mainly in the following three aspects:

[0004] Existing technologies primarily rely on equivalent sound pressure level (such as SPL) or ecological soundscape index (such as NDSI) to transform the originally intangible "soundscape" into a set of calculable perception standards for the precise planning and management of dynamic soundscape maps. However, the soundscape maps generated by these indicators can only answer "how loud is the sound here, and what percentage of it is natural sound," but cannot answer "whether people feel comfortable or annoyed here, or whether they would like to stay," resulting in dynamic soundscape maps that lack interpretability and application value.

[0005] Existing methods primarily incorporate static spatial features extracted from GIS and remote sensing imagery, such as road density and green coverage, as explanatory variables when constructing model training datasets. Because this type of data lacks dynamic temporal attributes, it can only output the long-term average soundscape of urban space and cannot respond to real urban rhythms such as rush hours and holidays. Consequently, its results are insufficient to support refined management scenarios requiring dynamic perception.

[0006] Current technologies heavily rely on a limited number of fixed monitoring stations and manual on-site recording for sample collection. However, such samples are sparsely covered in space and severely insufficient in quantity, resulting in a limited number of samples and high costs, making it difficult to support large-scale, high-resolution soundscape modeling.

[0007] In summary, in order to build a refined and high-fidelity dynamic soundscape map system, it is urgent to solve the three key problems in existing technologies: the single soundscape perception index, the static data system, and the high cost of sample acquisition. Summary of the Invention

[0008] To address the shortcomings of existing technologies, such as static nature, single-dimensionality, limited coverage, and coarse-grained representation, this invention provides a method, system, and medium for constructing dynamic soundscape maps of urban spaces.

[0009] To achieve the above objectives, the present invention provides a method for constructing a dynamic soundscape map of urban space, comprising:

[0010] The study collected subjective perception scores of residents in multiple areas of the target city on the regional soundscape quality, and simultaneously collected objective environmental data for each area. The objective environmental data included acoustic data, ecological soundscape index, and street view visual features within the area. Using subjective perception scores as the dependent variable and objective environmental data as the independent variable, the comprehensive soundscape perception index U-SDI equation was constructed using the generalized linear model GLM.

[0011] The target city is divided into multiple grid units according to a fixed spatial scale; static spatial data on building form, road network and green plant distribution, as well as dynamic environmental data on traffic flow and pedestrian density are acquired; sound sources in the dynamic environmental data are first simulated in batches, and then the simulated data is corrected to generate a sound pressure level (SPL) correction distribution field; the SPL correction distribution field data, ecological soundscape index and street scene visual features are substituted into the U-SDI equation to obtain the U-SDI value of each grid, and the U-SDI value is used as the ground truth label; static spatial features and dynamic environmental features of static spatial data and dynamic environmental data are extracted respectively, and the static spatial features and dynamic environmental features are integrated into a feature set according to the grid unit, and matched one by one with the ground truth label of the corresponding grid to form a soundscape training sample set.

[0012] Using the soundscape training sample set as input and U-SDI ground truth labels as output, a LightGBM model is trained to obtain a LightGBM model that can output predicted U-SDI values ​​for constructing dynamic soundscape maps of urban spaces.

[0013] Preferably, after the LightGBM model outputs the predicted value U-SDI, the method further includes: mapping the predicted U-SDI value onto each raster cell in a GIS platform, and generating a dynamic urban spatial soundscape map after spatial processing and visualization rendering.

[0014] Preferably, the acoustic data includes equivalent sound pressure level (SPL) and spectral structure; the ecological soundscape index includes at least one of normalized difference soundscape index (NDSI), acoustic complexity index (ACI), and sound entropy index (Ht); the street view visual features include at least one of sky openness, proportion of natural elements, and street enclosure degree extracted from street view images through a deep learning model.

[0015] Preferably, the static spatial characteristics include at least one of the following: green space coverage, water body coverage, artificial surface ratio, road density, building coverage, average building height, street enclosure, land use entropy, point of interest (POI) entropy and distance to high-grade roads; the dynamic environmental characteristics include at least one of the following: traffic flow, pedestrian density and time variables within different spatial scales.

[0016] Preferably, the step of first simulating the sound sources in the dynamic environmental data in batches, and then correcting the simulated data to generate a sound pressure level (SPL) correction distribution field specifically includes: using acoustic simulation software to generate an initial sound pressure level (SPL) distribution field in batches based on the dynamic environmental data; setting up a limited number of measurement points in the target city to obtain measured SPL data; and constructing a deviation correction model by comparing the measured SPL data with the simulated SPL data, and correcting the initial SPL distribution field to obtain the corrected SPL distribution field.

[0017] Preferably, the spatial processing includes neighborhood smoothing and boundary preservation processing, and the visualization rendering adopts a hierarchical color scheme.

[0018] Preferably, the method further includes: when a user triggers a target grid cell, the dynamic soundscape map calls a trained LightGBM model, and through the embedded SHAP method, decomposes and visualizes the contribution of static spatial features and dynamic environmental features to the U-SDI prediction value of the target grid; based on the contribution, it identifies the dominant factors that cause the soundscape quality to be good or bad, and then generates a targeted soundscape optimization strategy.

[0019] This invention also provides a system for constructing dynamic soundscape maps of urban spaces, comprising:

[0020] The data acquisition module is used to collect residents' subjective perception scores of regional soundscape quality in multiple areas of the target city, and simultaneously collect objective environmental data of each area; the objective environmental data includes acoustic data, ecological soundscape index and street view visual features within the area; with subjective perception scores as the dependent variable and objective environmental data as the independent variable, the comprehensive soundscape perception index U-SDI equation is constructed through the generalized linear model GLM.

[0021] The sample construction module is used to divide the target city into multiple grid units according to a fixed spatial scale; acquire static spatial data of the target city regarding building form, road network, and green plant distribution, as well as dynamic environmental data of the target city regarding traffic flow and pedestrian density; firstly, batch simulations are performed on the sound sources in the dynamic environmental data, and then the simulation data is corrected to generate a sound pressure level (SPL) correction distribution field; the SPL correction distribution field data, ecological soundscape index, and street scene visual features are substituted into the U-SDI equation to obtain the U-SDI value of each grid, and the U-SDI value is used as the ground truth label; the static spatial features and dynamic environmental features of the static spatial data and dynamic environmental data are extracted respectively, and the static spatial features and dynamic environmental features are integrated into a feature set according to the grid unit, and matched one by one with the ground truth label of the corresponding grid to form a soundscape training sample set.

[0022] The model training module is used to train the LightGBM model with the soundscape training sample set as input and U-SDI ground truth labels as output, so as to obtain a LightGBM model that can output predicted values ​​of U-SDI for constructing urban spatial dynamic soundscape maps.

[0023] The present invention also provides a computer-readable storage medium storing a computer program that, when loaded by a processor, can execute any of the steps in the method for constructing a dynamic soundscape map of urban space.

[0024] The method for constructing a dynamic soundscape map of urban space provided by this invention has the following beneficial effects:

[0025] This invention integrates subjective perception ratings with multi-source objective data, constructing a comprehensive soundscape perception index (U-SDI) that aligns with human experience through a generalized linear model (GLM). This calibrates the model's learning objective to real human perception, solving the problems of single-dimensionality and insufficient interpretability of the index. Employing acoustic simulation and limited experimental correction techniques, it efficiently generates U-SDI ground truth labels covering a high-resolution grid across the entire city, achieving dense full-area coverage of training samples at extremely low cost, overcoming the cost and coverage limitations of relying on on-site sampling. Simultaneously, it integrates static spatial features characterizing the city's framework with dynamic environmental features reflecting its operational pulse (such as traffic flow and pedestrian density), enabling the model to learn the spatiotemporal variation patterns of soundscapes. The LightGBM model trained on this high-quality dataset ultimately outputs accurate and interpretable dynamic soundscape predictions, upgrading the soundscape map from a static macroscopic background image into a high-fidelity dynamic monitoring and diagnostic system supporting refined governance and human-centered design. Attached Figure Description

[0026] To more clearly illustrate the embodiments and design schemes of the present invention, the accompanying drawings required for this embodiment will be briefly described below. The drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1 This is a flowchart illustrating a method for constructing a dynamic soundscape map of urban space according to an embodiment of the present invention.

[0028] Figure 2 This is a flowchart of the method for constructing a comprehensive soundscape perception index system according to an embodiment of the present invention;

[0029] Figure 3 This is a flowchart of the dynamic soundscape prediction model construction method according to an embodiment of the present invention;

[0030] Figure 4 A flowchart is provided for the dynamic soundscape map generation and soundscape optimization strategy of this invention. Detailed Implementation

[0031] To enable those skilled in the art to better understand and implement the technical solutions of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and should not be construed as limiting the scope of protection of the present invention.

[0032] This invention, through analysis, reveals that urban soundscapes are inherently influenced in real-time by highly dynamic factors such as traffic flow, pedestrian density, and time of day (e.g., morning rush hour and late night). Existing models, lacking effective input of this dynamic data, generate soundscape maps that are essentially "static snapshots" of specific time periods or average states. These models fail to accurately depict the dynamic characteristics of soundscapes as they change with urban operations, making it difficult to support real-time or near-real-time soundscape management and intervention. Current soundscape quantification indicators are fundamentally limited to physical acoustics and ecoacoustics data. While these objective indicators can describe sound source composition and energy distribution, they cannot fully characterize the multidimensional attributes of soundscapes at the "human perception" level, such as feelings of pleasure, tranquility, and a sense of event. The failure to integrate multimodal information directly reflecting human psychological feelings, such as subjective questionnaires and streetscape visual environments, leads to discrepancies between soundscape evaluation results and residents' actual experiences, resulting in insufficient explanatory power and adaptability when guiding people-centered urban environmental design. Obtaining high-quality training samples heavily relies on fixed physical monitoring stations and manual on-site recording. This approach has inherent drawbacks such as a limited number of sampling points, uneven spatial coverage, and high manpower and time costs. The sparsity of the samples directly limits the training effect of machine learning models, resulting in limited generalization ability. It is difficult to maintain high-precision prediction capabilities in complex and diverse urban landscapes, and it cannot support the needs of large-scale, high-resolution modeling.

[0033] Due to the aforementioned limitations of sparse samples and data systems, existing soundscape maps typically employ coarse spatial grids. While these low-resolution maps can macroscopically reflect the urban soundscape pattern, they fail to capture soundscape differences at the block, street, and even building scales. Consequently, they are ill-suited to provide direct and actionable guidance for refined governance scenarios such as urban design, public space renovation, and traffic organization optimization.

[0034] Therefore, based on the above analysis, this invention proposes the following technical system. This system consists of four parts: "Construction of Comprehensive Soundscape Perception Indicators," "Generation of High-Coverage, Low-Cost Soundscape Samples," "Construction of Soundscape Prediction Model Based on Machine Learning," and "Generation and Display of High-Resolution Dynamic Soundscape Maps," forming a complete process chain from the construction of human-centered perception indicators, data collection, model learning, to dynamic expression. The system as a whole uses a 50m×50m urban grid as a unified data organization basis, integrating and managing acoustic information, street view visual features, urban static spatial attributes, and dynamic traffic and pedestrian flow data within the same spatial framework. By constructing a comprehensive soundscape perception index (Urban Soundscape Quality Index, U-SDI) that reflects human-centered experience as the core response variable, and utilizing the LightGBM model to learn the nonlinear relationship between multi-source environmental features and U-SDI, this invention achieves full-domain prediction of urban soundscape quality. Combining acoustic simulation, measured correction, and GIS spatial mapping methods, the system can ultimately present the urban soundscape status with high precision, wide coverage, and support for time-based updates, providing technical support for dynamic perception, time-period analysis, and refined governance.

[0035] This invention constructs a method for urban soundscape modeling and map generation that can dynamically perceive, accurately quantify human experience, controllable costs, and possess high spatial resolution through data acquisition, soundscape perception index system construction, U-SDI generation, LightGBM model training, dynamic data input, soundscape prediction, map rendering and visualization output, and diagnostic and optimization strategies. This method overcomes the shortcomings of existing technologies, such as staticity, one-dimensionality, limited coverage, and coarse-grained expression, thereby providing a reliable technical tool for the refined management and human-centered design of the urban sound environment.

[0036] Based on this, the present invention provides a method for constructing a dynamic soundscape map of urban space, specifically as follows: Figure 1 As shown, it includes:

[0037] S1. Collect residents' subjective perception scores of regional soundscape quality in multiple areas of the target city, and simultaneously collect objective environmental data for each area; the objective environmental data includes acoustic data, ecological soundscape index and street view visual features within the area; with subjective perception scores as the dependent variable and objective environmental data as the independent variable, construct the comprehensive soundscape perception index U-SDI equation through the generalized linear model GLM.

[0038] This invention first establishes a comprehensive soundscape perception index system that can truly reflect residents' perceptual experience, addressing the problem of soundscape evaluation being dominated by physical indicators and lacking subjective interpretability. To this end, this invention conducts questionnaire surveys and simultaneous acoustic sampling in representative urban spatial environments. Subjective evaluations are obtained by quantifying residents' perceptual dimensions such as pleasure, tranquility, sense of event, and comfort. Simultaneously, acoustic indicators such as equivalent sound pressure level (SPL) and spectral structure information are collected at the same location to extract ecological soundscape indices, such as the Normalized Difference Soundscape Index (NDSI), Acoustic Complexity Index (ACI), and Entropy of Soundscape (Ht). Furthermore, deep learning is used to automatically identify visual environmental features in street view images, such as sky openness, the proportion of natural elements, and street enclosure.

[0039] Based on the distribution characteristics of soundscape perception scores (subjective scores are typically ordered discrete values ​​of 0-10 or continuous values ​​of 0-100), this invention selects the corresponding generalized linear model (GLM). When the subjective score is a continuous value, a Gaussian GLM (with the link function being the identity function) is used; if it is an ordered discrete value, an ordered logistic GLM is used. The core assumption of the model is: the standardized resident subjective perception comprehensive score S (0-100 points) is used as the dependent variable (a higher score indicates better soundscape perception quality), and the key indicators of three categories of objective environmental data—acoustic indicators, ecological soundscape index, and street view visual features—are used as independent variables.

[0040] These multimodal features are paired one by one with subjective ratings in the grid space to form a perception sample set. Subsequently, this invention uses the subjective perception rating as the dependent variable and acoustic indicators, ecological soundscape index, and visual features as independent variables to establish a generalized linear model (Gaussian GLM) to estimate the statistical significance and weight coefficients of various indicators in the subjective perception formation process, generating a formula for calculating the comprehensive perception index U-SDI. U-SDI expresses residents' overall evaluation of soundscape quality on a uniform scale (e.g., 0-100) and serves as the target variable for the subsequent soundscape prediction model of this invention, ensuring that subsequent model training and spatial extrapolation always adhere to the "human-centered perception" framework. U-SDI, which is the subjective perception comprehensive rating S to be calculated, is expressed as follows:

[0041] ;

[0042] in, For the model intercept term, The i-th acoustic index after standardization includes the standardized value of the equivalent sound pressure level (SPL) and the spectral structure characteristic value, with a value of [-1, 1]. , which are the weighting coefficients for acoustic indicators. Positive coefficients indicate that the indicator improves U-SDI, while negative coefficients indicate that it suppresses U-SDI. The j-th standardized ecological soundscape index includes the standardized values ​​of NDSI, ACI, and Ht, with values ​​ranging from [-1, 1]. The weighting coefficients for the ecological soundscape index. The standardized k-th street view visual feature includes the standardized values ​​of sky openness, proportion of natural elements, and street enclosure, with values ​​ranging from [-1, 1]. denoted as the weighting coefficient of the street view visual features, where m, n, and p are the number of acoustic indicators, ecological soundscape index, and street view visual features after significance screening, respectively, all of which are greater than 1.

[0043] Taking a target city as an example, 100 representative areas (covering residential areas, commercial areas, parks, transportation hubs, etc.) were selected. 30 subjective perception questionnaires were distributed to each area, and 2,860 valid questionnaires were collected. The subjective perception comprehensive score S (standardized to 0-100 points) for each area was calculated. Acoustic data of each area were collected simultaneously: equivalent sound pressure level SPL (dB(A)), 63Hz-8kHz spectrum energy ratio; ecological soundscape index: NDSI (0-1), ACI (0-1000), Ht (0-1); street scene visual characteristics: sky openness (0-1), proportion of natural elements (0-1), street enclosure degree (0-1).

[0044] Table 1. Weighting coefficients and intercept terms obtained from fitting data for a target city.

[0045]

[0046] All objective metrics are Z-score standardized to obtain standardized metrics. , and Abnormal samples (such as extreme noise regions with SPL > 85 dB(A) and discrete regions with questionnaire score standard deviation > 2) were removed, leaving 92 valid region samples. A Gaussian GLM model was constructed with S as the dependent variable and 10 standardized objective indicators (SPL, spectral proportion (2 items); NDSI, ACI, Ht (3 items); sky openness, proportion of natural elements, street enclosure (3 items); totaling 8 items) as independent variables. Significant indicators were screened using stepwise regression (P < 0.05), ultimately retaining 6 significant indicators: SPL (… ), NDSI ( ), ACI ( ), proportion of natural elements ( ), street enclosure degree ( ) and the proportion of the 63Hz spectrum ( The specific calculated values ​​are shown in Table 1, where "+" indicates a positive number, "-" indicates a negative number, and blank items indicate that the indicator did not pass the significance test and was not included in the final model.

[0047] =50.23−12.67 +3.89 +8.54 +4.21 +9.76 -5.32 After normalization, this formula can be directly used to calculate the U-SDI value (0-100 points) of any grid cell in the target city. The larger the negative coefficient index (SPL, street enclosure), the lower the U-SDI and the worse the soundscape quality. The larger the positive coefficient index (63Hz spectrum percentage, NDSI, ACI, natural element percentage), the higher the U-SDI and the better the soundscape quality.

[0048] The coefficient of determination R between predicted values ​​and measured subjective scores 2 =0.78, and the root mean square error (RMSE) = 6.32 points, indicating that the U-SDI formula can effectively fit the residents' subjective perception scores, has good explanatory power, and can be used as the core basis for subsequent raster true value label calculation.

[0049] The comprehensive perception index maintains the explanatory basis of subjective perception while also possessing the scientific rigor and accessibility of the samples required for machine learning. Through this technical solution, soundscape evaluation is upgraded from a single physical sound level or a single ecological soundscape index to a human-centered perception index system that comprehensively considers "physical, ecological, visual, and psychological" factors.

[0050] S2. Divide the target city into multiple grid units according to a fixed spatial scale; acquire static spatial data on building form, road network and green plant distribution of the target city, as well as dynamic environmental data on traffic flow and pedestrian density of the target city; first, perform batch simulation of sound sources in the dynamic environmental data, and then correct the simulation data to generate a sound pressure level (SPL) correction distribution field; substitute the SPL correction distribution field data, ecological soundscape index and street scene visual features into the U-SDI equation to obtain the U-SDI value of each grid, and use the U-SDI value as the ground truth label; extract the static spatial features and dynamic environmental features of the static spatial data and dynamic environmental data respectively, integrate the static spatial features and dynamic environmental features into a feature set according to the grid unit, and match them one by one with the ground truth labels of the corresponding grid to form a soundscape training sample set.

[0051] To address the issues of high cost and insufficient spatial coverage in traditional soundscape sample collection, this invention constructs a large-scale, high-density soundscape training sample system by combining acoustic simulation with limited field measurements. Specifically, after acquiring urban three-dimensional spatial data (including building height and shape, road network, topography, etc.), this invention first uses GIS spatial analysis methods to extract urban static spatial features. Urban static space includes variables such as the ratio of green space, water bodies, and artificial surfaces around the grid, road density and road hierarchy, building coverage and height, street enclosure, and land use entropy. These static spatial features are obtained through processing geographic information data and remote sensing imagery, and then spatially integrated into 50m×50m units.

[0052] Simultaneously, this invention also requires acquiring the dynamic environmental characteristics of urban operations over time. The dynamic data originates from multi-source urban operation platforms, including population heat maps extracted from map software, road-level traffic flow data obtained through web crawling and traffic monitoring systems. To improve data reliability, this invention collects a small amount of measured traffic and pedestrian flow data at multiple typical road nodes and functional areas, and uses this data to calibrate the acquired dynamic data, enabling the dynamic environmental characteristics to more accurately reflect the true state of urban operation.

[0053] After acquiring static and dynamic environmental characteristics, this invention utilizes the acoustic simulation software CadnaA to perform batch simulations of the propagation process of different sound source types (road traffic, rail transit, etc.) in urban space, generating a sound pressure level (SPL) distribution field for the study period. The simulation results are then aggregated at a grid scale, allowing each 50m × 50m grid cell to obtain simulated SPL values ​​for one or more time periods. Considering the potential for systematic bias in acoustic simulations, this invention further selects a limited number of representative road types, functional zones, and typical spatial forms to deploy a limited number of measured points. By comparing the measured and simulated results, an SPL bias regression model or a piecewise correction model is constructed to correct the overall SPL, making the acoustic input data more accurate.

[0054] The corrected SPL data is integrated with the processed ecological soundscape index and street view visual features to form the dependent variable U-SDI for each raster. At the same time, the static spatial features and dynamic environmental features obtained from GIS, remote sensing, street view recognition, and urban operation data are uniformly integrated according to the raster index, so that each raster unit contains a dependent variable and a complete set of multi-source variable features, forming a high-coverage sample set for training the LightGBM model.

[0055] Through this technical approach, the present invention significantly improves the spatial density and coverage of training samples without significantly increasing the cost of on-site data collection. This results in a significant improvement in the model's predictive stability and generalization ability at the urban scale, achieving the goal of conducting soundscape simulation across the entire urban area based on a small amount of subjective perception data.

[0056] S3. Using the soundscape training sample set as input and the U-SDI ground truth label as output, train the LightGBM model to obtain a LightGBM model that can output predicted U-SDI values ​​for constructing dynamic soundscape maps of urban spaces.

[0057] After completing the construction of the Integrated Soundscape Perception Index (U-SDI) and the generation of high-coverage training samples, this invention further constructs a soundscape prediction model based on LightGBM to achieve automatic inference and dynamic updating of urban soundscape quality across the entire spatial domain. This technical solution mainly includes three aspects: feature system construction, model training and validation, and a dynamic prediction mechanism.

[0058] In terms of feature system construction, this invention uses the Integrated Soundscape Perception Index (U-SDI) as the dependent variable, and multi-source environmental features reflecting the urban spatial structure and operational status as the independent variables of the model, with a 50m×50m grid as the unified data unit. The independent variables consist of two main categories: static spatial features and dynamic environmental features.

[0059] Static spatial characteristics are primarily used to describe the spatial morphology, built environment, and land use structure surrounding the grid. Examples include: green space coverage (LCgreen_per), artificial surface ratio (LCartifi_per), water coverage (LCwater_per), the degree of disorder in multi-functional land use structure (LUEntropy), and total road density (RdAll) and major road density (RdMajor) representing road conditions. Built environment characteristics include building coverage (BCR), street coverage (SCR), average building height (ABH), and distances to high-grade roads (such as expressways and arterial roads) (DisRdMajor, DisRdExpress). Furthermore, the degree of disorder in point-of-interest (POI) structure (POIEntropy) is considered to reflect the complexity of regional commercial, leisure, and population activities. These variables characterize attributes such as spatial enclosure, sound absorption / reflection structures, and functional mixing, which have a long-term shaping effect on the acoustic environment.

[0060] Dynamic environmental features are used to characterize time-varying factors that change with the rhythm of urban operation. This invention introduces traffic and pedestrian flow data at different scales, including near-scale traffic flow within 50 meters (trafficflow_near_50m), mid-scale traffic flow within 200 meters (trafficflow_mid_200m), far-scale traffic flow within 500 meters (trafficflow_far_500m), pedestrian density (pedestrian_flow_density), and the time variable "hour" representing the diurnal rhythm. These dynamic features can effectively capture the real-time impact of different road grades and activity types on soundscape quality over time, enabling the model to characterize typical urban rhythm changes such as peak-hour noise increases and nighttime sound environment recovery.

[0061] By establishing a correspondence between timestamps and grid indices, the dynamic features of each time slice can be integrated with static spatial features into a complete model input vector. The model can learn how urban spatial structure and temporal changes work together to affect the soundscape perception quality within a unified grid framework, laying a precise feature foundation for subsequent dynamic prediction mechanisms.

[0062] In the model building and training phases, this invention selects the LightGBM algorithm based on Gradient Boosting Decision Tree (GBDT) as the core prediction framework. On the one hand, LightGBM adopts a tree structure based on histogram-based node splitting and leaf-growing, which can effectively handle high-dimensional spatial features with multiple nonlinear interactions in soundscape prediction tasks. On the other hand, it has high training efficiency and strong generalization ability under large sample conditions, making it suitable for large-scale modeling at the city scale. This invention uses the U-SDI obtained by each grid cell in the sample building phase as the training label, and the corresponding static spatial features and dynamic environmental features as input variables. The model is trained and its performance is evaluated by combining K-fold cross-validation and spatial partitioning validation. At the same time, key hyperparameters such as learning rate, maximum depth, and number of leaf nodes are optimized using methods such as grid search and Bayesian optimization, thereby balancing model accuracy and computational efficiency.

[0063] Regarding the dynamic prediction mechanism, this invention establishes a time-slice-based model invocation and update process. When the system receives new traffic flow, pedestrian density, or time period information, it writes the latest dynamic environmental features into the input vector of the corresponding raster and invokes the pre-trained LightGBM model across the entire raster area to recalculate the U-SDI prediction value of each raster unit under that time slice. For scenarios that only need to evaluate planning schemes or long-term average states, the model can make predictions based on static spatial features and dynamic features of typical time periods (such as weekday daytime averages). For application scenarios that need to reflect the rhythm of urban operation, the input data can be continuously updated at preset time intervals (such as 15 minutes) to generate soundscape prediction results that evolve over time.

[0064] Through the above technical solution, this invention extends U-SDI from a small number of sample points to a city-wide raster, realizing detailed spatial extrapolation and dynamic temporal response of sound scene quality, providing a reliable predictive basis for subsequent generation of high-resolution dynamic sound scene maps, analysis of time-period differences, and scenario intervention simulation.

[0065] After completing the construction of the U-SDI index and the training of the LightGBM-based soundscape prediction model, this invention further constructs a high-resolution dynamic soundscape map generation system, enabling the spatial expression and structured display of soundscape prediction results at the city scale. This system aims to transform the model output into geographically meaningful, visually appealing, and analyzable soundscape information, and on this basis, form a technical optimization strategy framework for urban governance.

[0066] This invention, centered on "human-centered perception" and "dynamic precision," constructs a complete technology from data input to governance output. Using a 50m×50m grid unit as a unified spatial basis, it integrates acoustic features, subjective evaluation, visual features, ecological soundscape index, static spatial features, and dynamic environmental features. A generalized linear model (GLM) generates the U-SDI comprehensive perception index, which is then used for LightGBM model training and dynamic data updates to achieve city-wide soundscape prediction. Combined with SHAP value interpretation and GIS visualization technology, a high-resolution dynamic soundscape map is generated. Finally, soundscape diagnosis outputs targeted optimization strategies. The entire diagram clearly presents the entire process of "data collection, index construction, model training, dynamic prediction, map generation, and diagnostic optimization." Its core objective is to overcome the limitations of traditional soundscape technologies—static, one-dimensional, and low-resolution—to achieve dynamic perception, refined expression, and scientific governance of urban soundscapes. The SHAP dominant factor determination adopts a dual threshold standard: ① If the absolute value of a single feature's SHAP value accounts for ≥30% of the sum of the absolute values ​​of all feature SHAP values, it is determined to be a single dominant factor; ② If there are 2-3 features with a proportion ≥20% and a difference ≤5%, it is determined to be a composite dominant factor. At the same time, the threshold is adjusted in combination with the grid spatial scale: the above standard is used for 50m×50m grids, and if it is expanded to 100m×100m grids, the single dominant factor threshold is lowered to 25% to ensure the consistency of the determination logic under different scales.

[0067] like Figure 2 As shown, this invention revolves around a comprehensive soundscape perception index system. First, subjective scores are obtained through a questionnaire survey. At the same time, acoustic indicators (such as SPL), ecological soundscape indices (such as NDSI, ACI, Ht), and visual features (such as the proportion of natural elements) are collected through field measurements and integrated into multimodal features. Then, the subjective scores and multimodal features are input into a generalized linear model (GLM) to finally calculate the comprehensive soundscape perception index U-SDI, where X represents various acoustic, visual, or ecological indicators, and w represents the corresponding weights. This fully presents the construction process of the comprehensive soundscape perception index.

[0068] like Figure 3 As shown, the complete process of the dynamic soundscape prediction model technology based on LightGBM in this invention is as follows: First, street scene visual features, ecological soundscape index, and SPL obtained by CadnaA acoustic simulation are integrated to obtain the dependent variable U-SDI; at the same time, static spatial features (green space coverage, building coverage, density of main roads) and dynamic environmental features (traffic flow, pedestrian density, etc.) are integrated using 50m×50m grid units as a unified carrier; then, these elements are input into the LightGBM model for training, and after training, the construction process of the dynamic soundscape prediction model and related feature analysis are fully presented through SHAP value interpretation and feature importance analysis.

[0069] like Figure 4 As shown, the complete process from dynamic soundscape prediction results to targeted optimization strategies is as follows: First, the dynamic environmental features are input into the LightGBM prediction model to obtain the U-SDI dynamic prediction results; then, after processing by the spatial processing module (smoothing, boundary preservation, and correction), the raster hierarchical rendering module generates soundscape levels I-V; subsequently, with the help of the SHAP dominant factor identification module, three types of scenarios are identified: traffic-dominated, insufficient greening, and high building density, and corresponding targeted optimization strategies are output for traffic flow control and detour design, vegetation reinforcement and vertical greening, and facade optimization and interface reconstruction.

[0070] In its overall approach, this invention uses a 50m×50m grid as a unified spatial carrier, mapping the U-SDI prediction values ​​generated by the model at different time slices to an urban soundscape database. Through methods such as spatial smoothing, boundary preservation processing, and hierarchical rendering, a soundscape layer with continuity, structural clarity, and temporal dynamics is constructed. Based on these layers, the system can intuitively display soundscape information at multiple spatial scales, from overall layout and street differences to road nodes.

[0071] During the map generation stage, this invention imports U-SDI values ​​into the GIS raster structure according to different time slices and performs necessary spatial processing to avoid visual noise caused by prediction fluctuations, while preserving key boundary structures such as roads, water bodies, and green spaces. After these processing steps, the system uses a hierarchical color scheme to present soundscape levels, for example: Level I: High perception quality area (quiet, strong sense of nature); Level II: Nature-dominated area (obvious ecological soundscape); Level III: Mixed sound area (coexistence of natural and artificial sounds); Level IV: Artificial sound-dominated area (strong traffic and human voices); Level V: Low perception quality area (strong noise disturbance), enabling users to quickly identify typical spatial structures such as traffic-dominated areas, mixed sound areas, and areas dominated by natural sounds.

[0072] At the dynamic display level, this invention constructs a dynamic sequence of soundscapes based on a time-slice switching mechanism, enabling the map to reflect the changing trends of soundscapes throughout the day and under different circumstances. Users can observe the evolution of soundscape states during peak traffic hours, quiet nighttime periods, or specific events through the timeline function, thereby understanding the fluctuation patterns of soundscapes with the rhythm of urban operation.

[0073] The core innovation of this invention comes from the introduction of "dominant factor identification" technology for low U-SDI areas. Leveraging the interpretability extension of the LightGBM model, this invention deeply embeds SHAP (Shapley Additive Explanations) values ​​into a dynamic soundscape map system, enabling users to perform "point-level" soundscape cause analysis on any grid cell on the map. When a user clicks on a grid cell with a low U-SDI, the system calls the single-point SHAP decomposition result from the prediction model to generate a feature contribution map for that point. This map displays the positive and negative impacts of features such as traffic flow intensity, green space ratio, road density, building enclosure degree, and pedestrian density on the soundscape at that point, sorted by contribution value.

[0074] By visualizing SHAP values ​​(such as waterfall plots or horizontal bar charts), the system can clearly identify: which factors are the main drivers of soundscape degradation (e.g., traffic_flow_near_50m contributes a strong positive impact); which factors have an improving effect on the soundscape but are insufficient in the spatial structure (e.g., LCgreen_per contributes negatively but with a small absolute value); and how the interaction effects of multiple factors work together to affect the current U-SDI value. This mechanism transforms "model predictions" into "interpretable soundscape diagnoses." Users do not need a machine learning background to directly understand the root causes of soundscape changes through the interactive interface. Compared to traditional sound environment assessment methods, this invention not only provides "whether the soundscape is good or bad," but also accurately points out "why it is good or bad," thereby significantly improving the pertinence and scientific rigor of soundscape management.

[0075] In the optimization strategy section, this invention uses SHAP dominant factor identification as the core basis to construct soundscape improvement paths for different scenarios. Spatially, it proposes strategies including traffic flow control in traffic-dominated areas, vegetation reinforcement in areas lacking green space, and interface optimization in areas with significant building reflections. Temporally, by analyzing the response intensity and recovery capacity of the soundscape at different time points, it proposes differentiated control strategies applicable to peak-hour management, nighttime sensitive area control, and management during special events, enabling optimization measures to accurately match the spatiotemporal variation characteristics of the soundscape.

[0076] Through the above methods, this invention realizes a complete technology chain from dynamic prediction to spatial representation and then to intelligent diagnosis and optimization strategy generation, making dynamic soundscape maps not only a display tool, but also the core technology for soundscape governance and urban operation management.

[0077] In summary, the technical solutions of this invention mainly involve the fields of dynamic prediction, data integration, spatial mapping and visualization of urban soundscapes, and propose a set of soundscape prediction and optimization technologies based on multi-source data fusion and machine learning.

[0078] 1. This invention integrates acoustic data, ecological soundscape index, street view visual features, and subjective perception data, and employs a generalized linear model (GLM) to construct a comprehensive soundscape perception index (U-SDI), which serves as the core dependent variable for dynamic soundscape prediction. This method merges environmental data with human experience, achieving a precise quantitative expression of urban soundscape quality. The method integrates acoustic, ecological, and visual features with subjective ratings, and uses GLM to construct a standardized comprehensive perception index (U-SDI).

[0079] 2. This invention generates high-density sound scene samples covering the entire domain by combining acoustic simulation with measured data. By correcting the simulation results, the accuracy of the data is improved, providing high-quality training data for subsequent sound scene prediction. The method of combining acoustic simulation with measured data correction generates high-coverage sound scene samples, improving sample density and prediction accuracy.

[0080] 3. This invention employs a LightGBM-based model, combining static spatial features and dynamic environmental features for global soundscape quality prediction. This model can handle high-dimensional data and effectively learn the nonlinear relationship between features and U-SDI. The application of the LightGBM algorithm, combining static spatial features and dynamic environmental features, enables high-precision soundscape quality prediction.

[0081] 4. This invention designs a time-slice-based dynamic input mechanism, enabling soundscape predictions to automatically update over time. When urban operating conditions change, the model recalculates the U-SDI based on new traffic flow, pedestrian density, and other data. Furthermore, this invention maps the U-SDI prediction values ​​to a 50m×50m grid and generates a high-resolution dynamic soundscape map using techniques such as neighborhood smoothing and boundary-preserving filtering, accurately showcasing soundscape differences within the city. The time-slice-based dynamic input mechanism and real-time update method, combined with high-resolution rasterized soundscape map generation technology and neighborhood smoothing and boundary-preserving processing, ensure map accuracy and readability.

[0082] 5. This invention constructs a GIS-based dynamic soundscape visualization platform, displaying the U-SDI distribution at different time periods through time-slice switching, sequential playback, and multi-scale zooming, achieving a dynamic presentation of the soundscape in accordance with the rhythm of urban operation. Based on this, this invention introduces an interpretable analysis mechanism based on SHAP, allowing users to click on any raster to view the feature contribution decomposition of U-SDI at that point, identifying dominant influencing factors such as traffic, green space, and building enclosure. Furthermore, this invention constructs a soundscape optimization strategy framework based on the dominant factor results, proposing corresponding technical improvement paths for traffic-dominated, insufficient green space, or reflectivity-enhanced areas. The invention includes: a temporal display method for dynamic soundscape maps; SHAP-based raster-level dominant factor identification; and a soundscape optimization strategy framework oriented towards dominant factors.

[0083] Compared with the closest existing technologies (such as the urban soundscape quality mapping framework based on random forest and SHAP proposed by Jingyi Wang et al.), this invention achieves significant improvements in data system, indicator construction, sample generation, prediction model capabilities, and spatial representation accuracy. It can reflect urban soundscape quality more comprehensively, dynamically, and closely resemble the real-world operation of cities. The main advantages are as follows:

[0084] (1) From static to dynamic: Introducing the dynamic prediction capability of urban operating rhythm.

[0085] Existing technologies primarily focus on static spatial features, lacking the ability to depict dynamic factors such as traffic flow, pedestrian density, and temporal variations, thus failing to reflect the true temporal changes in soundscapes as cities operate. This invention, for the first time, incorporates multi-scale traffic flow, pedestrian density, and temporal variables into the soundscape prediction system, establishing a dynamic prediction mechanism of "time slice-feature input-model update." This allows soundscape maps to be updated on a minute-by-minute or hourly basis, realistically presenting key temporal characteristics such as morning and evening peak hours. This dynamic prediction capability provides more practical technical support for urban operation management and time-of-day governance, something that existing static soundscape frameworks struggle to achieve.

[0086] (2) From single-dimensional to multi-modal: Constructing a more human-centered integrated soundscape perception index (U-SDI).

[0087] Existing technologies rely entirely on measured acoustic metrics (SPL, NDSI), failing to adequately represent the subjective perception dimension. This invention, based on questionnaire surveys and simultaneous acoustic data acquisition, integrates acoustic metrics, ecological soundscape indices, and streetscape visual features. It also constructs a comprehensive perception index (U-SDI) using GLM (Geometrical Soundscape Index), enabling soundscape quantification results to more closely approximate real-world human experiences. Compared to existing technologies, this invention's soundscape index not only enhances explanatory power but also provides a more reliable basis for urban design, psychological restoration, and perception optimization.

[0088] (3) From scarce samples to high coverage: Construct a low-cost, high-density soundscape training data system.

[0089] Existing technologies rely on a large number of recording points and monitoring stations, resulting in high costs and insufficient coverage. This invention generates high-coverage SPL data through CadnaA acoustic simulation. This allows for the construction of a comprehensive and structurally complete training sample system covering the entire region at extremely low cost. This invention significantly increases data density without increasing acquisition costs, enabling the model to maintain high-precision prediction capabilities at the city scale—something traditional methods relying on monitoring points cannot achieve.

[0090] (4) From traditional machine learning to higher performance models: LightGBM improves large-scale prediction capabilities.

[0091] Existing technologies using random forests for urban soundscape prediction, while achieving a certain level of accuracy, still suffer from insufficient efficiency in large-scale urban modeling and high-dimensional data processing. This invention employs LightGBM based on GBDT, capable of handling higher-dimensional, multimodal, nonlinear, and strongly interactive features. Furthermore, it improves training and prediction efficiency through histogram acceleration and leaf growth strategies, making the model more suitable for city-level massive raster extrapolation while achieving higher prediction accuracy and stability.

[0092] (5) From macro to fine: The spatial resolution of soundscape maps has been significantly improved.

[0093] Existing soundscape maps generally use a coarse-scale grid of 100m×100m, which is insufficient to guide practical governance at the street and node levels. This invention uses a uniform high-resolution grid of 50m×50m and employs methods such as boundary smoothing and neighborhood consistency processing to enable the generated dynamic soundscape map to accurately represent the microstructure of road nodes, street spaces, and green space distribution, significantly enhancing its application value in urban renewal, street renovation, and ecological design.

[0094] (6) From “Results Display” to “Cause Diagnosis”: Introduce SHAP’s raster-level dominant factor identification function.

[0095] Existing technologies only use SHAP for variable importance analysis at a macro scale, failing to achieve rapid diagnosis of single-point soundscapes. This invention deeply embeds SHAP values ​​into a dynamic map, allowing users to click on any grid cell to obtain a decomposition of the feature contribution of that point (e.g., is traffic dominant? Is green space insufficient?). This invention elevates soundscape prediction results from "quantification" to "interpretable diagnosis," providing management departments with immediate evidence of "where the soundscape problems are, why they are poor, and how to improve them," achieving a point-level interpretive depth that traditional models struggle to reach.

[0096] (7) From static analysis to strategy generation: constructing an optimization path framework based on dominant factors.

[0097] Existing technologies typically only focus on soundscape evaluation. This invention further constructs a soundscape optimization strategy framework based on the SHAP dominant factor results, providing targeted improvement paths for different types of soundscape degradation mechanisms. For example, for traffic-dominated systems, it implements traffic flow control and detour design; for insufficient greenery, it implements vegetation reinforcement and vertical greening; and for high building reflectivity, it implements facade optimization and interface reconstruction. This transforms soundscape maps from "expression tools" into "governance tools."

[0098] Based on the same inventive concept, this invention also provides a system for constructing dynamic soundscape maps of urban spaces, comprising:

[0099] The data acquisition module is used to collect residents' subjective perception scores of regional soundscape quality in multiple areas of the target city, and simultaneously collect objective environmental data of each area; the objective environmental data includes acoustic data, ecological soundscape index and street view visual features within the area; with subjective perception scores as the dependent variable and objective environmental data as the independent variable, the comprehensive soundscape perception index U-SDI equation is constructed through the generalized linear model GLM.

[0100] The sample construction module is used to divide the target city into multiple grid units according to a fixed spatial scale; acquire static spatial data of the target city regarding building form, road network, and green plant distribution, as well as dynamic environmental data of the target city regarding traffic flow and pedestrian density; firstly, batch simulations are performed on the sound sources in the dynamic environmental data, and then the simulation data is corrected to generate a sound pressure level (SPL) correction distribution field; the SPL correction distribution field data, ecological soundscape index, and street scene visual features are substituted into the U-SDI equation to obtain the U-SDI value of each grid, and the U-SDI value is used as the ground truth label; the static spatial features and dynamic environmental features of the static spatial data and dynamic environmental data are extracted respectively, and the static spatial features and dynamic environmental features are integrated into a feature set according to the grid unit, and matched one by one with the ground truth label of the corresponding grid to form a soundscape training sample set.

[0101] The model training module is used to train the LightGBM model with the soundscape training sample set as input and U-SDI ground truth labels as output, so as to obtain a LightGBM model that can output predicted values ​​of U-SDI for constructing urban spatial dynamic soundscape maps.

[0102] The present invention also provides a computer-readable storage medium storing a computer program that can be used to execute the above-described method for constructing a dynamic soundscape map of urban space.

[0103] Specific limitations on the computational system for constructing dynamic soundscape maps of urban space can be found in the above-mentioned limitations on the construction method of dynamic soundscape maps of urban space, and will not be repeated here. Each module in the aforementioned system for constructing dynamic soundscape maps of urban space can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0104] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification. Furthermore, the above embodiments only illustrate several implementation methods of this application, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make several modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for constructing a dynamic soundscape map of urban space, characterized in that, include: The study collects subjective perception ratings of the soundscape quality of multiple areas in the target city from residents, and simultaneously collects objective environmental data for each area. The objective environmental data includes acoustic data, ecological soundscape index, and street view visual features within the area. The street view visual features include at least one of sky openness, proportion of natural elements, and street enclosure degree extracted from street view images of multiple areas in the target city through a deep learning model. Using subjective perception ratings as the dependent variable and objective environmental data as the independent variable, a comprehensive soundscape perception index (U-SDI) equation is constructed using a generalized linear model (GLM). The target city is divided into multiple grid units according to a fixed spatial scale; static spatial data on building form, road network, and green plant distribution, as well as dynamic environmental data on traffic flow and pedestrian density, are acquired; sound source data of the dynamic environmental data are acquired, the sound source data is simulated in batches, and the simulated data is corrected to generate a sound pressure level (SPL) correction distribution field; the SPL correction distribution field data, ecological soundscape index, and street scene visual features are substituted into the U-SDI equation to obtain the U-SDI value of each grid, and the U-SDI value is used as the ground truth label; static spatial features of the static spatial data and dynamic environmental features of the dynamic environmental data are extracted, and the static spatial features and dynamic environmental features are integrated into a feature set according to the grid units, and matched one by one with the ground truth labels of the corresponding grids to form a soundscape training sample set; Using the soundscape training sample set as input and U-SDI ground truth labels as output, a LightGBM model is trained to obtain a LightGBM model that can output predicted U-SDI values ​​for constructing dynamic soundscape maps of urban spaces.

2. The method for constructing a dynamic soundscape map of urban space according to claim 1, characterized in that, After the LightGBM model outputs the U-SDI predicted value, it also includes: mapping the U-SDI predicted value onto each raster cell in the GIS platform, and generating a dynamic urban spatial soundscape map after spatial processing and visualization rendering.

3. The method for constructing a dynamic soundscape map of urban space according to claim 1, characterized in that, The acoustic data includes the equivalent sound pressure level (SPL) and spectral structure; the ecological soundscape index includes at least one of the normalized difference soundscape index (NDSI), acoustic complexity index (ACI), and acoustic entropy index (Ht).

4. The method for constructing a dynamic soundscape map of urban space according to claim 1, characterized in that, The static spatial characteristics include at least one of the following: green space coverage, water body coverage, artificial surface ratio, road density, building coverage, average building height, street enclosure, land use entropy, point of interest (POI) entropy, and distance to high-grade roads; the dynamic environmental characteristics include at least one of the following: traffic flow, pedestrian density, and time variables within different spatial scales.

5. The method for constructing a dynamic soundscape map of urban space according to claim 1, characterized in that, The process of acquiring sound source data from the dynamic environmental data, performing batch simulations on the sound source data, and correcting the simulated data to generate a corrected sound pressure level (SPL) distribution field specifically includes: using acoustic simulation software to generate an initial SPL distribution field based on the dynamic environmental data through batch simulations; deploying a limited number of measurement points in the target city to acquire measured SPL data; constructing a deviation correction model by comparing the measured SPL data with the simulated SPL data, and correcting the initial SPL distribution field to obtain the corrected SPL distribution field.

6. The method for constructing a dynamic soundscape map of urban space according to claim 2, characterized in that, The spatial processing includes neighborhood smoothing and boundary preservation processing, and the visualization rendering adopts a hierarchical color scheme.

7. The method for constructing a dynamic soundscape map of urban space according to claim 1, characterized in that, Also includes: When a user triggers a target grid cell, the dynamic soundscape map calls the trained LightGBM model and uses the embedded SHAP method to decompose and visualize the contribution of static spatial features and dynamic environmental features to the U-SDI prediction value of the target grid. Based on the contribution, the dominant factors that cause the quality of the soundscape are identified, and then targeted soundscape optimization strategies are generated.

8. A system for constructing a dynamic soundscape map of urban space, characterized in that, include: The data acquisition module is used to collect residents' subjective perception scores of regional soundscape quality in multiple areas of the target city, and simultaneously collect objective environmental data of each area. The objective environmental data includes acoustic data, ecological soundscape index, and street view visual features within the area. The street view visual features include at least one of sky openness, natural element proportion, and street enclosure degree extracted from street view images of multiple areas of the target city through a deep learning model. With subjective perception scores as the dependent variable and objective environmental data as the independent variable, a comprehensive soundscape perception index (U-SDI) equation is constructed using a generalized linear model (GLM). The sample construction module is used to divide the target city into multiple grid units according to a fixed spatial scale; acquire static spatial data of the target city regarding building form, road network, and green plant distribution, as well as dynamic environmental data of the target city regarding traffic flow and pedestrian density; acquire sound source data of the dynamic environmental data, perform batch simulation of the sound source data, and correct the simulated data to generate a sound pressure level (SPL) correction distribution field; substitute the SPL correction distribution field data, ecological soundscape index, and street scene visual features into the U-SDI equation to obtain the U-SDI value of each grid, and use the U-SDI value as the ground truth label; extract the static spatial features of the static spatial data, extract the dynamic environmental features of the dynamic environmental data, integrate the static spatial features and dynamic environmental features into a feature set according to the grid units, and match them one by one with the ground truth labels of the corresponding grids to form a soundscape training sample set; The model training module is used to train the LightGBM model with the soundscape training sample set as input and U-SDI ground truth labels as output, so as to obtain a LightGBM model that can output predicted values ​​of U-SDI for constructing urban spatial dynamic soundscape maps.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is loaded by the processor, it is able to perform the steps of the method according to any one of claims 1 to 7.

Images