Study On Forecasting Model Of Land Subsidence And Its Application

Land subsidence is a main geologic disaster for a city,causing envioment quality problems which impair basal facilities'estabilishment and restrict economical sustainable development.It is said that land subsidence is mainly caused by groundwater overexploitation unbalancing groundwater in geologic dynamics. Taking Tanggu district in Tianjin for the study area ,a exact and applied subsidence prediction model is established which is significant to prevent and control geologic disaster, achieve sustainable utilization of groundwater resource and effectively control land subsidence.Spatial cluster is employed during the optimization of land subsidence monitoring network. Annual volumetric extraction from different confined aquifer and annual average water levels are introduced as model input, while land subsidence rate of optimized monitoring points is defined as model output.A multiple input and multiple output neural network model is established to achieve land subsidence's parallel prediction.A multiple regression modle is also established. The modeling results are proofed to be statistically significant which guarantees the explanation ability of the models, and the contour map of the neural network model outputs fit reasonably well with the spatial characteristics of land subsidence in the study area.According to model estimation, the average background land subsidence rate of Tanggu district is 9.47 mm a. Land subsidence and water levels of monitoring points respond basically the same to variation of extraction from certain aquifer. The significance of contribution of aquifers to land subsidence descends in order of theⅣ,Ⅲ,Ⅴ,Ⅱconfined aquifer.When hypothetic yield of groundwater meets 8 .46×106m3a, 1 .69×107m3a and 2 .70×107m3a, the corresponding estimated average land subsidence rate is 16.8 mm a, 30.9 mm a and 45.7 mm a respectively.To conquer the local optimization and the slow rate of cinvergence in BP neural network,the plobal-searching functions of geneti algorithms is used.It optimizes the initial weights of neural network and achieves the coupling of the genetic algorithms and BP neural networkThis algorithm is feasible and effective.Aided by GIS technology,land subsidence geographical information and prediction system is established and a human-computer interactional interface is designed.Procedures are executed at background.It can efficiently manage the information of water levels, exploitation and land subsidence of monitoring points and punctually fit the models.Input the exploitation or water levels data,push the button and the land subsidence prediction of the 22 monitoring points will be got.