| 基于随机森林模型的北京市五环内噪声地图模拟 |
| Noise Map within Beijing's Fifth Ring Road Based on Random Forest Model |
| 投稿时间：2023-11-17 修订日期：2024-03-19 |
| DOI：10.19316/j.issn.1002-6002.2024.04.25 |
| 中文关键词: 噪声 噪声地图 随机森林 预测 |
| 英文关键词:noise noise map random forest model prediction |
| 基金项目:国家自然科学基金(42071433) |
| 作者 | 单位 | | 刘宜婷 | 中国疾病预防控制中心环境与健康相关产品安全所, 中国疾病预防控制中心环境与人群健康重点实验室, 北京 100021 | | 白煜 | 南京航空航天大学经济与管理学院, 江苏 南京 211106
中国环境保护产业协会, 北京 100045 | | 许怀悦 | 中国疾病预防控制中心环境与健康相关产品安全所, 中国疾病预防控制中心环境与人群健康重点实验室, 北京 100021 | | 王情* | 中国疾病预防控制中心环境与健康相关产品安全所, 中国疾病预防控制中心环境与人群健康重点实验室, 北京 100021 | | 李湉湉 | 中国疾病预防控制中心环境与健康相关产品安全所, 中国疾病预防控制中心环境与人群健康重点实验室, 北京 100021 |
|
| 通讯作者:王情* 中国疾病预防控制中心环境与健康相关产品安全所, 中国疾病预防控制中心环境与人群健康重点实验室, 北京 100021 |
| 摘要点击次数: 94 |
| 全文下载次数: 107 |
| 中文摘要: |
| 基于声环境常规监测数据与道路交通、兴趣点、自然与社会经济、气象、空气质量、虚拟变量等特征参数,根据随机森林(RF)算法参数重要性排序结果,采用滑动窗口序贯向前选择法(SWSFS)进行参数选择,构建RF模型,预测北京市五环内100 m×100 m网格的噪声强度,绘制噪声地图,评估噪声时空分布特征,探讨影响噪声分布的主要因素。结果表明:2019年,北京市五环内声环境常规监测站点监测到的噪声强度为56.71 dB (A)±9.83 dB (A);采用RF模型预测得到的昼夜加权噪声强度为59.87 dB (A)±6.41 dB (A)，且深夜噪声低于白天和晚上。十折交叉验证结果表明,该模型预测性能较好,决定系数(R2)为0.78,均方根误差(RMSE)为4.65 dB (A)，平均绝对误差(MAE)为3.60 dB (A)。相比土地利用回归模型(LUR)，RF模型更优,其R2提高了35.09%，RMSE和MAE分别降低了24.13%和23.46%。RF模型特征参数重要性排序结果显示,道路交通(尤其是交通繁忙的主要道路)、兴趣点(尤其是公交车站、餐饮场所、购物场所)以及时间段等是影响噪声分布的主要因素。RF模型可以作为反映北京等特大城市噪声情况的一种可靠方法,为噪声暴露评估提供有效手段。 |
| 英文摘要: |
| By combining environmental noise routine monitoring data with road traffic, points of interest, natural and social economic factors,meteorology,air pollution,time variables,etc.,a random forest (RF) model was used to create noise maps and evaluate the spatiotemporal distribution characteristics of noise within Beijing's 5th Ring Road. The sliding windows sequential forward selection (SWSFS) was used to select the optimal parameter set. The model's variable importance ordering was used to explore the influencing factors of noise. The results showed that the noise intensity detected by the routine monitoring of the acoustic environment within Beijing's 5th Ring Road was 56. 71 dB (A) ± 9.83 dB (A) in 2019. The weighted noise intensity predicted by RF model was 59.87 dB(A) ±6.41 dB(A), and the noise in late night was lower than in day and night. The model validation results indicated that the model had good predictive performance,with a ten-fold cross-validation R2 of 0.78,root mean square error (RMSE) of 4. 65 dB (A), and mean absolute error (MAE) of 3. 60 dB (A). Compared with the Land Use Regression (LUR) model,the R2 of the RF model increased by 35. 09%, whereas the RMSE and MAE decreased by 24. 13% and 23. 46%, respectively. The RF model's variables importance ordering showed that road traffic, especially main roads with busy traffic flow,points of interest,especially bus stops,restaurants,shopping places,and time periods,were the main factors affecting the spatial variation of noise. The RF model can be a robust method for reflecting noise variability in megacities such as Beijing and may provide an efficient solution for noise exposure assessment. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
|
|
|
