基于多源数据的城市慢行系统声景品质评价与预测

Evaluation and Prediction of Soundscape Quality in Urban Pedestrian Street System Based on MultisourceData

关键词:  风景园林  慢行系统  声景品质  多源数据  BP神经网络  预测模型

Key words:  landscape architecture  pedestrian street  soundscape quality  multi-source data  BP neural network  prediction model

摘要:

城市慢行系统是城市公共空间的重要组成部分，其声景品质直接影响人们的游览体验。基于福州市鼓楼区19条慢行街道获取多源数据，从声级与声源、 空间形态、视觉环境和功能业态4个维度探讨潜在影响慢行系统声景的主要因素，构建城市慢行系统声景品质预测模型并对比预测结果。结果表明：1)声景愉悦 度与15个指标显著相关，影响较大指标包括声源变异量均值、谈话声和谐度；2)声景丰富度与11个指标显著相关，影响较大指标包括绿地形状指数、谈话声优势 度；3)声景品质BP神经网络预测模型整体拟合优度均大于90%，预测精度高于线性回归预测模型。

Abstract:

Under the context of healthy cities and the improvement of urban spatial quality in China, the urban pedestrian street system, as a vital public space for daily life and social interaction, has garnered increasing attention. The quality of their soundscapes directly influences people's recreational experiences and quality of life. Therefore, how to enhance the quality of the acoustic environment in the urban pedestrian street system through scientific soundscape design has become a significant topic in urban planning and landscape design area. In recent years, research has explored the influencing factors of urban pedestrian street system soundscapes from multiple dimensions. However, traditional research methods primarily rely on questionnaires and field monitoring, which often have limited sample sizes and struggle to comprehensively consider the effects of multidimensional factors. With the rapid advancement of technology, the application of multi-source data in soundscape research has gradually increased, providing new approaches for the evaluation and prediction of soundscape quality. Additionally, artificial intelligence technologies, such as BP neural networks, have made significant progress in soundscape prediction. This study examines 40 monitoring points on 19 streets in the urban pedestrian street system of Gulou District, Fuzhou. Using multi-source data, including remote sensing images, street view images, POI data, and field monitoring, the study explores the factors influencing soundscape evaluation across dimensions like sound level and sources, spatial morphology, visual environment, and functional formats. A soundscape quality prediction model for the urban pedestrian street system is developed and compared. Data collection includes questionnaires on socio-demographic characteristics, sound source evaluation, and overall soundscape evaluation. Objective metrics like equivalent continuous sound pressure level (LAeq), foreground sound mean (L10), and background sound mean (L90) are measured using a sound level meter, with sound source variability mean (L10-L90) derived from L10 and L90 differences. Remote sensing images capture road and building elements, while street view images, taken at 1.6 meters, are semantically segmented to extract visual features. POI data are obtained via an API platform. Data analysis involves SPSS 26.0 and EXCEL for correlation and linear regression, MATLAB for BP neural network prediction, and Origin 2022 for model comparison. Correlation results indicate that the pleasantness of the urban pedestrian street system is significantly correlated with 15 indicators, among which the sound source variability mean (L10-L90) and the harmonious degree of conversation have a substantial impact. Eventfulness is significantly correlated with 11 indicators, with the green space shape index and the dominant degree of conversation being particularly influential. Weight analysis reveals the importance of various indicators for pleasantness and eventfulness. Results show that the sound level and source indicators have a significant impact on pleasantness, with L10-L90, harmonious degree of conversation, and harmonious degree of vehicle horn sounds being particularly important. Visual environment indicators, such as green view index and street height-towidth ratio, also significantly affect pleasantness. For eventfulness, the green space shape index, dominant degree of conversation sounds, and functional mix are highly important, indicating that the shape of green spaces and the functional mix around the urban pedestrian street system significantly influence eventfulness. Prediction results demonstrate that BP neural networks have certain applicability in urban pedestrian street system soundscape research, with the constructed soundscape prediction model outperforming linear regression predictions. The single-layer neural network prediction results show good fitting performance. To enhance the soundscape quality of the urban pedestrian street system, designers can strengthen sound source control and design, and optimize the surrounding spatial environment. This study provides new insights into the evaluation and prediction of urban pedestrian street system soundscape and offers targeted and actionable recommendations for improving soundscape quality from multiple dimensions.