ARCHITECTURE OF INTWEL ANALYZER

It is useful to apply intelligent technology for well log data analysis. Intelligent well log analysis system called IntWeL Analyzer, which is shown in Figure 1, has been developed, IntWeL Analyzer consists of eight modules, including data warehouse, on line analysis processing (OLAP), data mining, knowledge acquisition, visualization, inference engine, knowledge base and interface.

In IntWeL Analyzer first well log data is transferred into data warehouse, then it can be handled by on line analysis processing (OLAP module). Under data warehouse supporting, data mining module will extract knowledge from log data^. Knowledge acquisition module will convert data mining results into knowledge base. Inference engine module deploys knowledge to analyze well log data supported by knowledge base^. Users can interactive with the IntWeL Analyzer through interface. We propose the object-oriented knowledge representation for the system. The knowledge will be represented in frame and semantic network based on object-oriented technology. This approach has all the featiires of an objectoriented mechanism, such as encapsulation, inheritance and message processing. The system puts all production rules into method slots. In order to provide a powerful inference mechanism, as well as the maximum flexibility and convenience, the system proposes a high-level language, that is Inference Control Language (ICL), which can be used to describe knowledge and rules, and control the inference process.

INDUCTIVE LEARNING

Inductive learning is an attractive research area in machine learning currently. The set of production rules generated by inductive learning is understandable to humans and this is one important reason we use it for the purpose of reservoir characterization. Among various inductive learning algorithms Ripper (Repeated Incremental Pruning to Producing Error Reduction) is one of the most effective and efficient. In this section we introduce Ripper algorithm first and then use it to analyze the well logging data in order to identify the pay zones of oil. Ripper algorithm proposed by Cohen in 1995^^. The underpinning of Cohen's algorithms is a descendant of REP (Reduced Error Pruning) which is a technique used in conjunction with a rule learning system in order to improve the accuracy of a generated rule set^V The whole algorithm of Ripper is consisted of two phases: the first is to determine the initial rule set and the second is post-process rule optimization.