Multiple Seismic Attributes, Neural Network Target Prediction Applied Research

The seismic multi-parameter goals prediction based on neural network is a kind of intelligence processing and explain method. It is used to describe the known and unknown samples and their relation indicating objective characters by Seism, Log and geological datum. The key of this kind of prediction is to set up the corresponding relation between the seismic characteristic parameter and geological parameter.The article use BP neural network prediction technology has realized that the synthetical prediction of many geological objectives, on the basis of analyzing and expounding the principles the neural network and seismic multi-parameter extracting method. At the same time, a multi-parameter geological objective prediction system based on BP neural network was developed. This system is developed by Borland C ++ Builder6.0, Visual Basic6.0, Matlab6.5 and Fortran Station4.0. It is mixed programming use the four kinds of languages. This system included six major modules, and the six major modules divide to twenty-six child module. The major modules: target stratum extract, transform processing, parameter extract, graphical display, prediction analysis and text editing. The whole systematic function is relatively overall and adaptability is relatively strong.The article is extracting the wavelet domain parameter by using the tool of wavelet transform, besides extracting general seismic parameter. The seismic characteristic parameters have nine sorts and 51 kinds. These are correlation characteristic parameter, Fourier spectrum characteristic parameter, power spectrum characteristic parameter, time domain amplitude characteristic parameter, linear prediction coding coefficients, instantaneous characteristic parameter, absorb and decay coefficient, velocity characteristic parameter and wavelet packet transform characteristic parameter. The parameters contain the surface relatively wide, the prediction which is suitable for the goal of many kinds of Seism needs.In parameter analysis predict module includes the normalization method and principal component analysis, etc. Make the neural network training efficiency and improve the speed.In order to real-time displaying the parameters and the predicts result, developed the parameters curve displaying module and the black-and-white / colored Seismic sections displaying module, which provide convenient conditions for the comparative choice of many kinds of parameters and outputting the result.At last, utilize developed the system to examine and predicted three different objectives: the theory model datum, the concrete member defect and coal layer thickness, The result is correct, it proves that the system is more available and practical in predicting of seismic multi-parameters.