CN 11-5366/S     ISSN 1673-1530
“风景园林,不只是一本期刊。”
邬峻. 第四范式:基于机器学习的荷兰智慧城市宜居性预测模型研究[J]. 风景园林, 2020, 27(5): 11-29. DOI: 10.14085/j.fjyl.2020.05.0011.19
引用本文: 邬峻. 第四范式:基于机器学习的荷兰智慧城市宜居性预测模型研究[J]. 风景园林, 2020, 27(5): 11-29. DOI: 10.14085/j.fjyl.2020.05.0011.19
WU Jun. The Fourth Paradigm: A Research for the Predictive Model of Livability Based on Machine Learning for Smart City in The Netherlands[J]. Landscape Architecture, 2020, 27(5): 11-29. DOI: 10.14085/j.fjyl.2020.05.0011.19
Citation: WU Jun. The Fourth Paradigm: A Research for the Predictive Model of Livability Based on Machine Learning for Smart City in The Netherlands[J]. Landscape Architecture, 2020, 27(5): 11-29. DOI: 10.14085/j.fjyl.2020.05.0011.19

第四范式:基于机器学习的荷兰智慧城市宜居性预测模型研究

The Fourth Paradigm: A Research for the Predictive Model of Livability Based on Machine Learning for Smart City in The Netherlands

  • 摘要: 本研究系统地调查了“第四次工业革命”背景下大数据爆炸给传统研究范式带来的挑战和机遇。以荷兰智慧城市宜居性预测模型为例,引入数据密集型的“第四范式”新研究方法。首先从传统研究范式和宜居性研究现状出发,收集与人居环境宜居性相关的变量和可用的多源数据集。然后执行必要的数据清理、数据工程、数据特征提取方面的工作流程,使收集到的原始数据能够满足机器学习的基本要求。随后在机器学习的反复实验中,选用监督式机器学习关于多级目标预测的2个通用算法:多类别决策丛林和多类别决策森林,经过比选和优化得到最优算法。然后,将这种最优算法部署到云计算环境中生成智能预测工具箱,以监测和预先干预荷兰人居环境的宜居性。研究表明,与传统研究范式相比,尖端的人工智能技术和基于“第四范式”开发的新兴机器学习算法确实在知识发现、定量分析、知识快速更新和预测研究中具备独特优势。该范式在处理未来智慧城市中更大量、多样、精准、快速的研究数据集时将更加高效并更具创新前景。

     

    Abstract: This research systematically investigated the challenges and opportunities brought by the big data explosion to the traditional research paradigms in the context of the fourth industrial revolution. Taking the predictive model of livability for Dutch smart cities, which had been developed through machine learning, as an example, the research introduced “the fourth paradigm” as a novel data-intensive research method. The research discussed the traditional research paradigms and the status of livability research, collected applicable variables and available multi-source big data sets related to the livability of human settlements. Afterward, the research proceeded with necessary data cleansing, data engineering, and feature engineering to ensure that the collected raw data sets met the basic requirements of machine learning. In the subsequent machine learning experiments, the research selected two general algorithms, namely Multiclass Decision Jungle and Multiclass Decision Forest, to carry out a multiclass prediction in supervised machine learning. These were compared and optimized to obtain an algorithm with higher accuracy. This optimized algorithm was deployed to the cloud to produce a smart forecasting toolbox to monitor, predict, and perform an early intervention on the livability of the Dutch environment. The research demonstrated that cutting-edge AI algorithms and emerging machine learning technologies developed on the basis of “the fourth paradigm” have a pronounced advantage in knowledge discovery, quantitative analysis, rapid knowledge update, and predictive research, as compared with the traditional research paradigms. The proposed paradigm is more efficient, innovative, and capable of dealing with research containing data sets with large volume, more variety, veracity, and velocity for future smart cities.

     

/

返回文章
返回