基于大数据和熵权-随机森林的城市地下空间需求评价

科学评估地下空间开发需求潜力是缓解城市化问题和合理拓展有限区域的重要基础工作。目前地下空间评价中的社会经济数据多来自于传统官方文件, 其全面完整性和时空精度并不理想; 此外主客观赋权方法的使用, 一定程度上存在主观性强和受数据干扰等不足。文章以多源大数据支持的指标体系为基础, 构建熵权-随机森林耦合的地下空间需求评价模型。该模型基于熵权法确定负样本, 将总样本和指标因子导入随机森林算法中, 挖掘社会经济指标与现有地下设施间的复杂非线性关系。研究表明, 经过网格搜索调优后的模型 AUC(area under curve) 精度达到 0.979, 其中 77.45% 的现有设施落入评价的高需求区内, 证明所采用模型有较强的准确性和可靠性, 其精细化评价结果可为今后地下建设选址提供更符合实际的借鉴。

Scientific assessment of underground space development demand potential is an important basic work to alleviate urbanization problems and reasonably expand limited areas. The current socioeconomic data in underground space evaluation are mostly from traditional official documents, and their comprehensive completeness and spatio-temporal accuracy are not ideal. In addition, the use of subjective and objective assignment methods is to a certain extent subjective and subject to data interference. This paper constructs an entropy-random forest coupled underground space demand evaluation model based on the index system supported by multi-source big data. The model is based on the entropy weight method to determine the negative sample, and the total sample and index factors are imported into the random forest algorithm to explore the complex nonlinear relationship between socio-economic indices and existing underground facilities. The case study shows that the area under curve (AUC) accuracy of the model after grid search tuning reaches 0.979, in which 77.45% of the existing facilities fall into the high demand area of the evaluation, which proves that the adopted model has strong accuracy and reliability, and its refined evaluation results can be used as a practical reference for future underground construction site selection.