About us
IPLab LLC is a software development company, resident of Skolkovo Innovation Center in Moscow, Russia, focusing on development of software products for prediction of oil and gas formations productivity parameters.
We solve the following tasks, using machine learning algorithms and next-generation Kolmogorov neural networks:
Analysis of the seismic field without a teacher (without wells)
- Building maps of seismofaces (seismic classes) based on classification applying 1D,2D or 3D Kohonen neural networks with RGB visualization.
- Selection of faults and fractures based on the DTW algorithm
- Tracing of faults based on the new algorithm of simulation of faults and disturbances with the use of multiple local stresses
- Selection of factors based on the orthogonal decomposition of the seismic field.
- Selection of seismic field features based on RGB images of adjacent stratigraphic slices of the seismic field or attributes
Analysis of the seismic field with a teacher (teacher - information from wells)
Prediction effective thickness maps.
- based on linear regression,
- based on non-linear ACE regression,
- based on Random Forest nonlinear regression,
- based on Kohonen neural networks
- on the basis of classical neural networks
- based on new generation Kolmogorov neural networks
- multiple forecast with removal of part of wells and construction of maps P10, P50, P90, average, standard deviation
Prediction the cube effective parameters based on a set of source cubes and well measurements.
- based on linear regressions,
- based on classical neural networks
- based on new generation Kolmogorov neural networks
- Neural network forecast of the low-frequency model and its application.
- Nonlinear prediction of cubes of elastic parameters AI, Vp/Vs, RHOB by angular sums
- Forecast of FES cubes by inversion cubes or angular sums
- Forecast of geomechanical cubes (speed Vs, young's modulus, Poisson ratio ...)
- Pore pressure cube forecast
- Forecast of lithofacies cubes
- multiple forecast with the removal of part of wells and construction of cubes P10, P50, P90, average, standard deviation
Analysis of well data.
Prediction of curves (core parameters) for set of logs
- based on linear regression,
- based on classical neural networks
- based on new generation Kolmogorov neural networks
- multiple forecast with the removal of part of wells and construction of curves P10, P50, P90, average, standard deviation
Classification of logging curves according to their shape-electrofacing
IPLab software is based on:
- A new generation of Neural networks based on Kolmogorov full function neurons with innovative methods of hybrid training.
- Seismic inversion applying full, angle and azimuth stacks.
- Fracture analysis based on seismic data set.
We use modern innovative machine learning algorithms to process and interpret complex data of various scales and different accuracy (well surveys, seismic exploration, ground research, aerospace surveys). The development of technologies based on such algorithms involves the use of large volumes of input data in order to build reliable forecasts of reservoir productivity for traditional and unconventional hydrocarbon deposits.
As the main algorithm for forecasting productivity, new-generation neural networks based on Kolmogorov neurons with full-featured activation functions are used, which will ensure a high degree of freedom of the nonlinear forecast operator. Training of neural networks and their stabilization is based on a combination of genetic algorithms, gradient methods and regularization. This method allows using multi-scale and multi-accuracy data at the input to forecast effective parameters of oil and gas production. In addition, we use several special techniques (know-how) for such a forecast related to the use of the spatial distribution of the initial data (distribution of the seismic field around the forecast point) and taking into account the multifactorial nature of the production data. Based on our experience, this significantly improves the forecast quality.
We use new ideas on seismic inversion based on "direct" inversion constructions in the spectral domain for full, angular and azimuthal sums. For this, an innovative theory of inversion constructions in the spectral domain is used. Our existing experience in implementing the White Wave Inversion technology (Petrel plugin available on the Ocean Store) and our new ideas in this area (there is a working prototype and patent applications are being prepared) allowed us to build a qualitatively new innovative technology of seismic inversions that can effectively compete in the world inversion market with well-known technologies of such companies as GeoSoftware (Jason Geoscience Workbench, Hampson-Russell), Schlumberger.
Special artificial intelligence algorithms are used to identify fracturing zones and hidden faults based on 3D seismic data. The use of machine learning algorithms allows us to solve such a problem much more effectively. We are currently preparing a patent application and have a working prototype of the plugin. Comparison of the results of identifying fracture zones and hidden faults based on various technologies and our approach shows greater resolution and efficiency.
The project is focused on the Russian and global market of geological and geophysical modeling and interpretation of seismic data. Our main customers are geological and geophysical modeling departments in oil and gas companies, service companies, research institutes and universities.