فهرست نشریات علمی دسترسی آزاد ایران

This paper is mainly about creating a novel method for detecting potential subsurface structures in the study area for future investigation. This paper is divided into two sections. In the first section, for notifying the statistical status of data points, two types of support vector regression are proposed. In the second section, by using a fuzzy inference system, potential parts of the study area are detected. For this research, 812 magnetic surveying points were collected in the southern part of Iran with a range of magnetic between. -105.718 to 2.20 two types of SVR are proposed, and in both methods, firstly, an MLP neural network predicts magnetic rates by using easting, northing, and elevation, and in MLP_SVR by trial and error main parameters of SVR  are chosen but in MLP_PSO_SVR optimized parameters of SVR  are chosen by the PSO algorithm. Main parameters of support vector regression are epsilon, penalty term, and kernel function (in this study we choose RBF kernel function), In MLP_SVR, epsilon is 24, penalty term is 27 and sigma of RBF kernel is 12, and optimized parameters in second type are estimated 28.27 for epsilon, and 21.99 for penalty term, and 15.70 for sigma of RBF kernel,, and by this parameters, SVR is performed. In the second section, utilizing unique FIS platform is created, and parameters of this intelligent system are predicted magnetic rate, which is predicted by an MLP neural network, and depth. To obtain depth, Standard Euler deconvolution is offered, which uses least-squares as an inversion method. MLP_SVR puts 753 data points inside model and MLP_PSO_SVR puts 775 data points inside model which means MLP_PSO_SVR has 2.70 % better performance in comparison of  MLP_SVR, in second section, the condition of the subsurface structure is defined, and the outcome of the model illustrates that in this area, if the magnetic rate is more than -80 and Simultaneously, the depth is less than 2000, that parts are proper area for future investigation. Main finding of this study are; (1) MLP_PSO_SVR causes that more data points in comparison of MLP_SVR be inside of Support vector regression model and this proposed type has better performance in comparison of MLP_SVR (2) particle swarm particle is great tool for optimizing main parameters of support vector regression (epsilon, penalty term, and sigma of kernel function) and The function of norm(sin⁡(x)) causes that choosing of optimized parameters of support vector regression becomes attainable   (3) fuzzy inference system creates novel procedure for notifying status of subsurface structure which can be used in future research in study area.