城镇建设适宜性评价的贝叶斯网络机器学习方法

Bayesian Network Machine Learning for Urban Construction Suitability Evaluation

研究目的:避免城镇建设适宜性评价指标及其权重体系建立的主观性,通过“数据海选—非线性认知—线性规则转化”的方式,探索能反映城镇建设适宜性影响因素之间真实联系、评价结果易于理解、便于落地实施的城镇建设适宜性评价方法。研究方法:非线性机器学习、贝叶斯网络模型。研究结果:贝叶斯网络非线性结构学习出的有向无圈网络(DAG)揭示了城镇建设适宜性评价指标影响的线性主次性,贝叶斯网络非线性参数学习出的各评价指标重要度映射出城镇建设适宜性评价指标的线性重要度。研究结论:对比“专家经验线性评价”、“非线性贝叶斯网络评价”“转化线性规则评价”三种方法评价结果,非线性贝叶斯网络评价结果的合理性大大高于专家经验线性评价结果,转化线性规则评价结果的准确性不如非线性贝叶斯网络评价,但误差在可接受范围内,评价出的城镇建设适宜用地更加集中连片,评价结果易于感知、理解,可行性强。

The purpose of this paper is to explore the urban construction suitability evaluation method from“data selection”,“nonlinear cognition”to“linear rule transformation”,which could reflect the real relationship among the influencing factors,to easily understood the results,and to avoid the subjectivity about selecting the indicators and their weights.The research methods include machine learning and Bayesian network model(BN).The results show that the BN structure learning can find the primary-secondary relationship among the influencing factors through Directed Acyclic Graph(DAG);and the BN parameters learning can reveal the importance of indicators from a linear perspective.In conclusion,comparing among expert experience linear evaluation,nonlinear Bayesian network evaluation and transformed linear rule evaluation,the rationality level of the result from nonlinear Bayesian network evaluation is significantly higher than that of expert experience linear evaluation,while the accuracy level of the result from transformed linear rule evaluation is lower than that of nonlinear Bayesian network evaluation with an acceptable error.However,using the linear rule transformed from BN Learning to evaluate the urban construction suitability can provide the rational results for concentrated and scale urban construction land use,which is easy to be understood and perform.