Soft computing for qualitative and quantitative seismic object and reservoir property prediction. Part 1: Neural network applications
Fred Aminzadeh and Paul de Groot of dGB Earth Sciences begin a major series of three articles on the increasing use of soft computing techniques for E&P geoscience applications, focusing first on how neural networks can enhance seismic object detection. Soft computing has been used in many areas of petroleum exploration and development. With the recent publication of three books on the subject, it appears that soft computing is gaining popularity among geoscientists. In this paper we focus on one aspect of soft computing: neural networks, in qualitative and quantitative seismic object detection. In subsequent papers we will review other aspects of soft computing in exploration. Highlighted here will be the role neural networks play in combining different seismic attributes and effectively bringing together data with the interpreter’s knowledge to decrease exploration risk in four categories (geometry, reservoir, charge and seal). Three new books in the general area of soft computing applications in exploration and development, Wong et al (2002), Nikravesh et al (2003) and Sandham et al (2003) represent a comprehensive body of literature on recent applications of soft computing in exploration. Soft computing is comprised of neural networks, fuzzy logic, genetic computing, perception- based logic and recognition technology. Soft computing offers an excellent opportunity to address the following issues: ■ Integrating information from various sources with varying degrees of uncertainty ■ Establishing relationships between measurements and reservoir properties ■ Assigning risk factors or error bars to predictions. Deterministic model building and interpretation are increasingly replaced by stochastic and soft computing-based methods. The diversity of soft computing applications in oil field problems and the prevalence of their acceptance can be judged by the increasing interest among earth scientists and engineers. Given the broad scope of the topic, we will limit the discussion in this paper to neural network applications. In subsequent papers we will review other aspects of soft computing, such as fuzzy logic in exploration. Neural networks have been used extensively in the oil industry. Approximately 10 years after McCormack’s review (1991) of neural network applications in geophysics, much work has been done to bring such applications to the main stream of geophysical interpretation. Some of these efforts are documented in Wong et al (2002), Nikravesh et al (2003) and Sandham et al (2003) which include many papers and extensive references on neural network applications. Most of these applications have been in reservoir characterization, seismic object detection, creating pseudo logs, and log editing. In the next section, we will focus on two general areas of applications of neural networks. This will include qualitative methods with the main aim of examining seismic attributes to highlight certain seismic anomalies without having access to very much well information. In this case neural networks are primarily used for classification purposes. The second category involves quantitative methods where specific reservoir properties are quantified using both seismic data and well data, and neural networks serve as an integrator of the information.