SeisPron

The first phase of program SeisSproN (2005-2013) was designed to use 2D projection (maps) of geophysical parameters ( reservoir properties - porosity , the effective thickness of the formation; well logs curves, etc. ) based on neural networks and linear regression. The calculation is performed for a set of seismic attributes, taken during the selected time window along the horizon and the predicted values ​​of the parameter, known at the locations of wells.

Geophysical parameters are calculated for the inter-well space in which pre-defined calculated seismic attributes. The program used an approach that allows using different neural networks get maps predictive parameter total (depending on all the measurements at the points of localization of holes), and the local character.

Stability prediction (the ability of the chosen model to calculate adequately result in new data) is determined using the procedures of cross-validation (cross-validation) and calculation errors , as well as by constructing cross- rafts and calculate the correlation coefficients obtained and the initial forecast parameters . The program allows you to perform clustering on a set of seismic attributes for seismic facies analysis . I

mplemented input and output seismic attributes and maps of parametrers for geophysical interpretation software - DV SeisGeo, DV- 1 Discovery, Petrel.

Valuing the attributes

To determine significant seismic attributes (those that contribute most significant contribution to the result), calculation used as attributes significance coefficients for the linear model (standardized coefficients of the linear regression) and on neural networks using special algorithms. Selecting only significant attributes not only help to reduce their number and simplify the calculations, but also simplifies the search for meaningful relationships between the original attributes and forecasted parameters. Calculation models

The program allows to find the linear and nonlinear relationship between seismic attributes and forecasted geophysical parameters and calculate forecast maps in accordance with different models. One of the main methods used to predict seismic attributes is a multiple linear regression . Using our software is easy to obtain linear relationships between the attributes and the forecast and build his card and make full regression analysis. However , in cases of non-linear functions, the program uses a more powerful approach based on the approximation of petrophysical relationships using neural network modeling. Neural networks

The program used an approach to get the maps of predictive parameters of general and local types. For general maps , depending on the measurements ( seismic and well data) at all points localization wells used neural network multilayer linear perceptron , and for the prediction of local type - generalized regression neural network . When learning