融合权重因子模型和深度学习方法的城市地面沉降危险性分析

摘要

Urban ground subsidence hazard induced by building load is analyzed and studied combined with weights of evidence model and deep learning method in the case of southeast subsidence areas,Tianjin,China.We discussed the value controlling or related to ground subsidence of seven major factors:building floor area ratio, structure form,basis form,slope,soil compression modulus,depth to groundwater and groundwater permeability based on weights of evidence model.we proposed the WOE-DBM model by combining the weights of evidence (WOE)with deep Boltzmann machine (DBM),which was applied to draw hazard index figure.The results were validated by receiver operating characteristic (ROC)which show the ground subsidence hazard index generated by this model has a certain "diagnostic"role on land settlement history case in the study area.The AUC is 0.83 that indicates prediction result coordinate with field survey data and certifies the model has high accuracy to ground sub-sidence hazard induced by building load assessment and prediction.The results can be widely used for hazard pre-vention,architecture pattern chosen and land-use planning in the densely urban areas.%以天津市东南部沉降区为例，结合权重因子模型和深度学习的方法对城市建筑群荷载作用下的地面沉降危险性进行了分析和研究。基于权重因子模型分析了研究区内建筑容积率、建筑结构和基础形式、地形坡度变化、土壤压缩模量、地下水的埋深和地下水渗透性七个方面的诱发因子对沉降危险性的影响大小，再根据WOE-DBM模型绘制了地面沉降与其诱发因子的危险性指数图。通过ROC检验表明，基于WOE-DBM模型生成的沉降危险性指数对研究区内已发生的沉降具有较好的“诊断”作用，AUC值达到了0.83，预测结果与实测结果也具有很好的一致性，从而证明该方法对于建筑物荷载引发沉降的评价和预测是非常有效的，其分析结果可以广泛应用于城市密集建筑区的地面沉降危害性预警、建筑形式选择以及城市规划的分析决策当中。