TY - GEN
T1 - Oil well performance modeling under measurement uncertainty using fuzzy logic
AU - Matthew, Olumuyiwa Oluwafunto
AU - Gegov, Alexander
AU - Abdulkadir, Abdullahi
AU - Ichtev, Alexandar
AU - Abdullahi, Saheed
PY - 2025/12/29
Y1 - 2025/12/29
N2 - Accurate reservoir characterization and hydrocarbon volume estimation are essential for efficient oil and gas exploration. This paper evaluates and compares the performance of a Mamdani type Fuzzy Inference System against several supervised machine learning algorithms (Random Forest, Support Vector Machine, Logistic Regression, and Naive Bayes) for reservoir zone classification. A well log dataset from the Kansas Geological Survey was analyzed, and the machine learning models achieved high classification accuracies, with Random Forest reaching 100%. The Fuzzy Logic system also demonstrated high accuracy (99%) while offering superior interpretability. Additionally, hydrocarbon volume was measured using Stock Tank Oil Initially In Place(STOIIP) equation, achieving a strong correlation (R2 of 0.97) with log derived parameters. The results confirm the effectiveness of both approaches, highlighting the trade-offs between the predictive power of machine learning and the expert-driven transparency of fuzzy logic, thereby aiding in more robust exploration decisions.
AB - Accurate reservoir characterization and hydrocarbon volume estimation are essential for efficient oil and gas exploration. This paper evaluates and compares the performance of a Mamdani type Fuzzy Inference System against several supervised machine learning algorithms (Random Forest, Support Vector Machine, Logistic Regression, and Naive Bayes) for reservoir zone classification. A well log dataset from the Kansas Geological Survey was analyzed, and the machine learning models achieved high classification accuracies, with Random Forest reaching 100%. The Fuzzy Logic system also demonstrated high accuracy (99%) while offering superior interpretability. Additionally, hydrocarbon volume was measured using Stock Tank Oil Initially In Place(STOIIP) equation, achieving a strong correlation (R2 of 0.97) with log derived parameters. The results confirm the effectiveness of both approaches, highlighting the trade-offs between the predictive power of machine learning and the expert-driven transparency of fuzzy logic, thereby aiding in more robust exploration decisions.
KW - reservoir characterization
KW - hydrocarbon volume estimation
KW - machine learning
KW - measurement uncertainty
KW - fuzzy logic
KW - well log data
UR - https://easychair.org/cfp/MMA2025
U2 - 10.1109/MMA67107.2025.11311204
DO - 10.1109/MMA67107.2025.11311204
M3 - Conference contribution
SN - 9781665478007
T3 - IEEE MMA Conference Proceedings
SP - 1
EP - 6
BT - 2025 XXXV International Scientific Symposium Metrology and Metrology Assurance (MMA)
PB - IEEE Xplore
T2 - MMA 2025
Y2 - 7 September 2025 through 11 September 2025
ER -