基于组合模型的天津市地面沉降预测及危险性评价

Combined model-based prediction and hazard assessment of land subsidence in Tianjin

针对天津市过度开采地下水引发的地面沉降问题,基于组合模型开展区域沉降预测及危险性评价研究。首先分别运用灰色理论GM(1,1)预测模型、BP神经网络预测模型以及灰色BP神经网络(GM-BP)组合预测模型,对地面沉降量数据进行校核与补充。然后选取高程、坡度、土地利用、河网水系、水文地质、地下水开采量和累积沉降量作为影响因子,基于确定性系数和逻辑回归组合模型对地面沉降的危险程度进行评价,最后将地面沉降区域划分为5类危险区:极高危险区、高危险区、中危险区、低危险区和极低危险区。结果表明:(1)灰色BP神经网络预测模型的稳定性和拟合能力明显优于其他两种预测模型,模型的预测值更能满足精度要求;(2)研究区不同程度的危险性评价中,41.57%的区域面积具有极高度和高度危险性,且主要分布于北辰区、津南区、静海区、西青区及滨海新区等南部地区;(3)通过灾害点与危险区叠加分析,整个研究区域灾害点密度为64.73处/万km^(2),45处灾害处于高和极高的地面沉降危险。分区结果与灾害点的分布情况基本吻合,证明该研究成果能为地面沉降预测及危险性评价提供参考依据。

Aiming at the problem of the land subsidence caused by groundwater over-exploitation in Tianjin,the study and evaluation of the regional land subsidence and hazard are carried out.At first,grey theory GM(1,1)prediction model,BP neural network prediction model and grey BP neural network(GM-BP)combined model are applied to the verification and supplement of the land subsidence data.And then,the elevation,slope,land-use,water system of river network,hydrogeology,volume of groundwater exploitation and cumulative subsidence are taken as the impacting factors to assess the hazard degree based on the certainty coefficient and logistic regression combined model.Finally,the land subsidence area is divided into five hazard regions,i.e.extremely high hazard,high hazard,mid-hazard,low hazard and extremely low hazard.The results show that(1)both the stability and fitting ability of the grey BP neural network model are obviously better than those of the other two prediction models and the predicting values of the model can better meet the relevant accuracy requirements;(2)in the assessment of different degrees of the hazard within the study areas,41.57%of the areas are under the conditions of extremely high hazard and high hazard and mainly distributed in the southern regions such as Beichen District,Jinnan District,Jinghai District,Xiqing District,Binhai New Area,etc.(3)through the superposition analysis of disaster points and hazardous areas,the density of disaster points in the whole study area is 0.65/10,000 km^(2),and 45 disaster points in the study area have high and extremely high hazards of land subsidence.The hazard zoning result is basically coincided with that of the actual distribution of the disaster points.It is proved that the study can provide a referential basis for the prediction and hazard assessment of land subsidence.