Potash Reservoir Prediction From Seismic Data Based On Convolution Extreme Learning Machine

Eastern Sichuan is rich in potash resources.In order to strengthen the exploration of potash resources in this area,under the guidance of the policy of " exploration of oil and potash",the research on potash reservoirs in eastern Sichuan has been carried out.Due to the lack of direct and clear relationship between potash content and seismic response,it is difficult to establish characteristic equation,so the current understanding of seismic exploration of potash reservoir is limited.In order to solve this problem,based on the seismic reflection characteristics and logging rock physical characteristics of potash reservoir,the paper conduct the study of potash reservoirs prediction based on the p-impedance inversion.Due to the limitation of strong seismic amplitude and resolution,the well coincidence rate of potassium salt reservoir prediction results based on p-impedance inversion results is not high,which can not meet the needs of further exploration of potassium salt resources.In view of the problem that it is unable to identify the potash reservoir directly and effectively based on the p-impedance inversion results,this paper studies the seismic potash reservoir prediction method based on the full connection network model in eastern Sichuan by using the strong nonlinear fitting ability of the full connection network.Firstly,based on the seismic attribute data and logging potassium content curve data,the supervised fully connected potassium salt reservoir prediction network is constructed,and then the potassium salt content data volume in East Sichuan area is directly obtained through the network,and then the distribution of potassium salt reservoir is predicted combined with the geophysical characteristics of potassium salt reservoir.Although this method can directly predict potash reservoir,due to the influence of network model,seismic characteristic attribute data sets and other factors,the generalization ability of the proposed fully connected potash reservoir prediction network is insufficient,which leads to the poor well coincidence rate of potash reservoir prediction results.In order to further improve the well coincidence rate of potash reservoir prediction network and reduce the recognition error of potash reservoir,according to the excellent generalization ability of extreme learning machine and the characteristics of convolution neural network self mining data features,this paper carried out the research on seismic potash reservoir prediction method based on convolution extreme learning machine in Eastern Sichuan.Firstly,based on the data of seismic wave field and logging potassium content curve,the convolution extreme learning machine potassium salt reservoir prediction network is constructed;Then,the data volume of potash content in eastern Sichuan is obtained directly through the network,and the distribution of potash reservoir is predicted combined with the geophysical characteristics of potash reservoir;Finally,the prediction method of potash reservoir based on p-impedance inversion and seismic potassium reservoir prediction method based on full connection network and convolution extreme learning machine are compared.The results show that the prediction results of potash reservoir by convolution extreme learning machine not only show good reservoir formation,but also have higher well coincidence rate.