Page 153 - Intelligent Digital Oil And Gas Fields
P. 153

116                                       Intelligent Digital Oil and Gas Fields



                                      Challenges
                                        Usage
                                        Quality
                                        Context
                                       Streaming
                                       Scalability





                      Data operators                Data modalities
                         Collect                      Ontologies
                         Prepare                      Structured
                        Represent                     Networks
                         Model                          Text
                         Reason                       Multimedia
                        Visualize                      Signals
          Fig. 4.6 Importance and relevant areas that interact in data mining projects. (Modified
          from Leskovec, J., Rajaraman, A., Ullman, J.D., 2014. Mining of Massive Datasets, second ed.,
          Cambridge University Press.)

          modeling and high-performance computing (HPC), modern data mining is
          largely extracting data models or patterns that can sometimes be the sum-
          mary of the data or even the set of most extreme features of the data.
          The data mining tasks are mainly classified into the following:
          •  Descriptive methods: where automated and intelligent tools discover pre-
             viously unknown human-interpretable patterns that describe the data.
             – Example: data clustering of reservoir parameters to identify sweet
                spots for new drilling campaigns (Roth et al., 2013).
          •  Predictive methods: where automated systems [e.g., recommendation sys-
             tem (Leskovec et al., 2014; Jordan and Mitchell, 2015)] and models use
             certain variables (predictors) to predict unknown future values, trends, or
             behavior of other (response) variables.
             – Example: well production prediction and optimization (Zhong et al.,
                2015) or equipment predictive maintenance, both based on histori-
                cally recorded data.
          •  Root-cause analysis: where automated tools are used to identify roots and
             causes of a system’s faults and problems, mostly based on the analysis of
             historical categorical, continuous, and temporal data.
             – Example: down-time or job-paused time analysis of hydraulic fractur-
                ing, well artificial lift, or well stimulation equipment (Maucec
                et al., 2015).
   148   149   150   151   152   153   154   155   156   157   158