Application of Machine Learning and Digital Music Technology to distinguish high from low gas-saturated reservoirs

In this paper, we discuss a novel approach of pattern recognition, clustering and classification of seismic data based on commonly applied techniques in the domain of digital music and in musical genre classification. Our workflow starts with accurate conversion of seismic data from SEGY to Musical Instrument Digital Interface (MIDI) format. Then, we extract MIDI features from the converted data. These can be singlevalued attributes related to instantaneous frequency and/or to the signal amplitude. Furthermore, we use multi-valued (or "high-level") MIDI attributes that have no equivalent in the seismic domain. For instance, we use MIDI features related to melodic, harmonic and rhythmic patterns in the data. We discuss an application to real data. We apply a Machine Learning approach to the MIDI-converted seismic data set with the purpose of accurate seismic facies classification. The final objective of the test is to distinguish between geological formations prevalently formed by clay, from two different gas-bearing sandy layers: one is a low gas-saturated reservoir and the other one is a high gas-saturated reservoir. In this paper, we present encouraging results. Considering the novelty of our approach, additional investigations are in progress on larger data sets, for a complete understanding of the physical meaning of the new "high-level" MIDI attributes.