|
| 基于机器学习的城市建成区噪声地图生成方法 |
| Generation of noise maps in urban built-up areas based on machine learning |
| 投稿时间：2025-04-20 修订日期：2025-06-23 |
| 中文摘要: |
| 交通噪声是威胁居民身心健康和生活质量的城市第二大污染源。噪声地图作为反映噪声分布的重要评估方法，其研究在成本控制、区域适应性或模型解释性方面仍面临挑战。本研究以大连市典型建成区为对象，提出快速评价交通噪声的模型方法。首先基于75m尺度网格单元整合遥感影像、空间句法、规划要素等多源数据；其次，构建基于网格单元内开源数据的机器学习模型，通过聚类将网格划分五类并赋予标签（如：枢纽区、绿地区、密路网小街区），分别建立回归模型预测噪声，绘制噪声地图；最后采用SHAP分析特征贡献机制，增强模型解释性。结果表明：（1）聚类与回归组合模型可解释76.3%的噪声变化，较线性模型性能提升49.0%。（2）高斯混合模型（GMM）聚类能有效识别空间异质性，随机森林（RF）回归的噪声预测精度和准确率较高。（3）通过SHAP提取关键特征，证明不同城区主导噪声分布的关键特征存在差异。研究有助于为精准降噪策略制定提供科学依据。 |
| 英文摘要: |
| Traffic noise is the second largest source of urban pollution that threatens the physical and mental health and quality of life of residents. As an important evaluation method to reflect noise distribution, noise mapping still faces challenges in terms of cost control, regional adaptability or model interpretability. This study takes a typical built-up area in Dalian as the research object and proposes a model method for rapid evaluation of traffic noise. First, based on the 75m grid unit, multi-source data such as remote sensing images, space syntax, and planning elements are integrated; secondly, a machine learning model based on open source data in the grid unit is constructed. The grid is divided into five categories by clustering and labeled (such as hub area, green area, and small block with dense road network). Regression models are established to predict noise and draw noise maps; finally, SHAP is used to analyze the feature contribution mechanism to enhance the interpretability of the model. The results show that: (1) The combined model of clustering and regression can explain 76.3% of the noise variation, which is 49.0% higher than the linear model. (2) Gaussian mixture model (GMM) clustering can effectively identify spatial heterogeneity, and random forest (RF) regression has high noise prediction accuracy and precision. (3) By extracting key features through SHAP, it is proved that the key features of the dominant noise distribution in different urban areas are different. The research helps provide scientific basis for the formulation of precise noise reduction strategies. |
| DOI： |
| 中文关键词: 交通噪声 噪声地图 机器学习 开源数据 城市建成区 空间异质性 |
| 英文关键词: Traffic noise Noise mapping Machine learning Open-source data Urban built-up areas Spatial heterogeneity |
| 基金项目:国家自然科学基金项目（面上项目，重点项目，重大项目）基于规划要素影响机制的交通噪声预估方法及规划管控研究 |
|
| 摘要点击次数: 6 |
| 全文下载次数: 0 |
|
查看/发表评论 下载PDF阅读器 |
| 关闭 |
