8: First Break Picking Using a Neural Network
Published:January 01, 1991
Traditional von Neumann computers using traditional programs are extremely good at number crunching tasks, but they cannot approach human performance in simple perceptual tasks such as recognizing a face or identifying a sound. This discrepancy, among other things, has been a major motivating factor in developing brain-based, massively parallel computing architectures. The neural net paradigm is one such paradigm that has proved to be good at pattern recognition tasks.
In exploration geophysics, the picking of seismic first arrivals represents a pattern recognition task. We attacked this problem by defining four signal attributes for each potential first arrival peak as input to a back propagation neural network. Then, by using a set of known (user selected) first arrival peaks, we trained the back propagation network to recognize first arrival peaks. The neural net based first arrival picking system achieved above 90 percent accuracy on picking several seismic surveys (each survey required a separate training).
Figures & Tables
Expert Systems in Exploration
“So far as the laws of mathematics refer to reality they are not certain; and so far as they are certain, they don’t refer to reality.” This quote from Albert Einstein explains the difficulties associated with a mathematical description of many subjective rules used by human experts in reaching a decision. This fact, combined with the multidisciplinary nature of oil exploration, has prevented wide spread use of expert systems in the oil industry. Alternative approaches, that can account for uncertainty and are better equipped for integrating data and rules from different disciplines, should be used to allow expert systems to become viable tools in our applications.
This book examines a diverse set of petroleum exploration problems that properly designed expert systems can help solve. Chapter 1 provides an extensive review of current state-of-the-art expert systems as pertains to oil industry problems. Emphasis is given to how uncertainty and inexactness of data and rules from different disciplines could be handled by expert systems. The chapter suggests that fuzzy logic, evidential reasoning, and neural networks will prove essential in the design of many expert systems that are capable of solving more practical exploration problems. The problem of automatic picking of stacking velocities, using a rule based system, is addressed in Chapter 2. The expert system automates the task of picking the extreme of the velocity spectrum. The system incorporates common sense rules to distinguish primary reflection related peaks in the spectrum from those related to multiples and noise.