| 摘要: |
| 【目的】大数据、物联网和人工智能技术正在经历快速发展阶段，其中机器学习的应用尤为瞩目，探索机器学习对可持续建成环境研究的影响具有理论和实践价值。【方法】基于文献综述，聚焦城市公共健康、能源碳排放、气候环境、生态系统、绿色出行5个可持续建成环境重要议题，详述机器学习的概念、分类、重要算法及关键应用。【结果】提出机器学习应用预测性有余解释力不足的特点，梳理机器学习发展从预测性到解释性的趋势，分析机器学习应用对研究的影响。【结论】结果表明：解释性方法和可读模型增多，研究目的更加侧重决策解读和规律总结，但基于实证研究的因果机制探索仍较少。基于此，比较分析了机器学习在不同议题中的典型应用，展望未来的发展前景。 |
| 关键词: 人工智能 解释性机器学习 公共健康 能源碳排放 气候环境 生态系统 |
| DOI：10.12409/j.fjyl.202303010108 |
| 分类号:TU984；TU986；TP181 |
| 基金项目:国家自然科学基金青年项目“基于人工智能的城市 PM2.5 时空格局”（编号 52108060）；国家重点研发计划“十四五重点研发子课题：形流匹配的国土空间综合效能评价”（编号 2022YFC3800804）；上海市自然科学基金项目“人工智能评估城市 PM2.5 浓度的时空格局及规划响应——以上海市为例”（编号 21ZR1466500）；上海市科技支撑双碳专项“面向区域碳中和的超大城市规划设计关键技术”（编号 22DZ1207800） |
|
| Application of Machine Learning in the Research on Sustainable Built Environment: Progress and Prospect |
|
LIU Zerun, LIU Chao
|
|
Tongji University
|
| Abstract: |
| [Objectives] The development of big data and the internet of things (IoT) has accelerated the process of artificial intelligence, with machine learning (ML) becoming particularly noticeable. More people are realizing the influence of advanced technology on reframing the researches in the field of sustainable built environment. Thus, this research aims to identify the role of machine learning in the research on sustainable built environment. [Methods] By reviewing the conceptions of sustainability and built environment, the research summarizes several important topics in the wide-ranging field of sustainable built environment. Based on literature from Web of Science (WoS) and China National Knowledge Infrastructure (CNKI) over the last five years, the research finds that the increasing trends of ML-related research are significant for topics such as urban public health, energy consumption and carbon emission, climate and environment, ecosystem, and green travel, which may serve as a useful clue for structure organization and conclusion drawing of this research. Initially, the research elaborates on the concept, classification and algorithms of ML, and presents the application characteristics of strong prediction capability but poor interpretation capability. Then the research analyzes the trend change of ML from predictability to interpretability, followed by the descriptions of scope, taxonomy and core algorithms of interpretable ML. Moreover, the research establishes a thorough table to list multiple ML algorithms and their roles in the research. Far from being all-inclusive, the research only tries to incorporate as many methods and objectives as possible across the world. The research particularly focuses on the predictability and interpretability of the results of ML algorithms, and the specific roles of such algorithms in the research direction, methodology and conclusion of sustainable built environment. [Results] The results show that ML’s interpretability capability substantially aids in the transformation of research techniques. More and more researchers are practicing ML-based research frameworks and building practical methodological routes, accompanied by increasingly diversified research directions and objects. ML has excellent performance in dealing with complex, heterogeneous, dynamic and large-scale data. Complicated issues previously considered difficult to solve have become the deciphering objects of ML, providing new insights and possibilities for solving composite puzzles. Meanwhile, ML itself is progressing toward intelligence, transparency and interpretability. By integrating experience and human wisdom, ML can help researchers formulate optimal decision strategies, reveal underlying mechanisms, and uncover universal laws. Furthermore, the research findings are more scientific and sophisticated. The majority of researches in the five key topics recognize the great advantages of ML in prediction and interpretation and start to develop interpretable and readable methods based on specific scenarios and problems. It is clear that more researches are focusing on understanding decision-making processes and mining patterns, although the levels of progress thereof vary depending on specific issues. As for public health, several articles discuss the effects of the built environment on health concerns. The applications of interpretable ML are relatively simple and preliminary, with less empirical investigation of transparent models. More medical and clinical experiments are needed to support theoretical analyses. As for energy consumption and carbon emission, as well as climate and environment, both predicting and interpreting tasks are important. By upgrading data processing, the accuracy of estimation and prediction can be improved. Additionally, models’ inner processes can be better understood by optimizing algorithms. As for ecosystems, there are more researches on prediction than on interpretation. The complexity of ecological elements might be the primary cause for this渠慳汩⁴捵桡慴湩捯敮猬†慷湨摩扨爠敭慡歹琠桢牥漠畣杨桡獮楥湤†牷敨汥慮琠敭摯晥椠数汯摷獥慦牵敬†汭楯此敥汬祳†瑡潲攠浣敲牥条整楤渮†瑁桳攠景畲琠畧牲敥n travel, ML applications have reached a mature stage. Lots of cutting-edge technologies have been developed and widely implemented. Except for the predictable and interpretable tasks, some researches have further stepped into the next level of intelligent decisions and optimal strategies, providing a vibrant foundation for human-machine interaction. [Conclusions] In summary, there is no doubt that ML plays an important role in transforming research methods, objects and conclusions in the field of sustainable built environment, and can effectively promote the development of researches in this field in a more quantitative, diverse, intelligent and scientific manner. It can be foreseen that the researches and applications of ML in sustainable built environment will continue to advance swiftly, with long-term and broad prospects. Besides, the shift from predictability to interpretability is obvious. On the one hand, there are more interpretable methods and readable ML models, and research objectives tend to emphasize understanding the decision-making process and revealing the underlying rules; on the other hand, the exploration of casual mechanisms based on clinical experiments is still insufficient. In terms of the five key topics, the development of ML research and application is the fastest in transportation issues such as green travel, forming a mature system of prediction, interpretation, and decision-making, while continuously promoting the exploration of more innovative applications and methods of human-machine interaction. There are also quite full prediction and interpretation tools for researches on energy consumption and carbon emission, and climate and environment. However, both the variety and professionalism of research methods remain insufficient. Though the ML applications in ecosystems and public health are still in the early stage, many researchers have realized the great potential of ML in these areas, implying that additio |
| Key words: artificial intelligence interpretable machine learning public health energy consumption and carbon emission climate and environment ecosystem |
