Detection Of Karst Cave Using Cross Hole Resistivity Method With Experience-based Learning Algorithm

Unfavorable geological conditions,such as groundwater,boulder,and karst cave,are usually existed in the karst area during constructio Cross hole resistivity method,one of the advanced geological prediction method,is often used to detect the underground geological conditions in advance.However,this method still has some shortcomings in the field of observation model,inversion theory,and model parameters.The sensitivity iterative method based on Tikhonov regularization is sensitive to initial value and noise,which makes it easy to fall into local optimum.Moreover,the traditional observation mode commonly possesses a great number of defects,such as small amount of observation data,low inversion accuracy,and unable to effectively suppress multiple solutions.in order to overcome these shortcomings and investigate the law of the influence between detection accuracy of cross hole resistivity method and model parameters(borehole spacing,electrode spacing,cave location and borehole depth to aperture ratio),the following work is done in this paper:1.A new quadrupole observation model is proposed.Three finite element numerical examples including a small cave,a large cave,and a complex cave are conducted and cave inversion results in different observation modes are and analyzed.The comparison results show that the new quadrupole observation mode can significantly increase the amount of observation data and improve the effectiveness of data,and hence enhance the accuracy of inversion.The new mode ultimately provides more accurate underground cave advance prediction results.2.An experience-based learning algorithm(EBL)is proposed.Three numerical examples of small,large and beaded caves,and three different water filled cave models(water filled,semi water filled and air filled)with three different shapes(including fissure,bag and cone)are conducted.The inversion imaging results by using different swarm intelligence search algorithms and sensitivity iterative method based on Tikhonov regularization are d analyzed.The results show that the EBL has fast convergence speed and high accuracy.The proposed method can clearly reflect the shape and water filling state of the cave,and can significantly enhance the inversion imaging resolution and accuracy of the cross hole resistivity method.3.The analysis of model parameters of cross hole resistivity method is carried out.Geoelectric model of karst cave with different model parameters are established to analyze the inversion imaging law of water filled karst cave and gas filled karst cave with different parameters.The results show that the detection accuracy of the cross hole resistivity method is inversely proportional to the borehole spacing and electrode spacing,and directly proportional to the ratio of borehole depth to pore diameter.Meanwhile,the inversion effect of the cave model is the best in the middle of the detection area.4.The laboratory physical model test of karst cave was carried out.The new quadrupole observation mode is employed for pole running measurement,and the collected current and potential difference data are imported into matlab program of the EBL algorithm.The comparison results between the inversion and the actual situation shows that the inversion results basically reflect the real location of the cave physical model under the influence of potential difference measurement error,indoor physical model error and background soil heterogeneity.The inversion method based on empirical learning algorithm and the actual inversion effects of the new quadrupole observation model are preliminarily verified.5.The field experiment of cave detection was carried out.The new quadrupole observation mode is used for pole running measurement,and the collected data is imported into matlab program of the EBL algorithm.The inversion and drilling results show that the inversion results can accurately reflect the location and shape of karst caves,and thus improve the accurate geological prediction results for practical engineering.The practical engineering application effects of the new observation model using the EBL algorithm are further verified...