Naoufal ACHARKI

Naoufal ACHARKI

Senior Data Scientist | Marketing ML Expert | Causal Inference Specialist

Causal Inference for Energy Production Optimization

Published:

Overview

Conducted advanced research at the intersection of statistical learning and causal inference for energy production optimization, developing novel methods for gas and geothermal well prediction under uncertainty.

Challenge

Energy companies face critical decisions with:

  • High uncertainty in well production forecasting
  • Need for robust prediction intervals, not just point estimates
  • Causal questions about operational interventions
  • Expensive field experiments requiring reliable predictions

Solution

Research Contributions

1. Gaussian Process Regression for Production Prediction

  • Developed robust prediction interval estimation methods
  • Achieved 80% confidence in gas well production forecasts
  • Implemented cross-validation techniques for interval calibration
  • Published in Computational Statistics & Data Analysis (CSDA) journal

2. Causal Inference for Geothermal Wells

  • Applied treatment effect estimation to operational decisions
  • Developed causal ML methods for continuous treatments
  • Compared meta-learners (T-learner, S-learner, X-learner, R-learner)
  • Published at ICML 2023 (top-tier ML conference)

3. Uncertainty Quantification

  • Bayesian modeling approaches for decision-making under uncertainty
  • Sensitivity analysis for model robustness
  • Integration of domain knowledge with statistical methods

Technology Stack

  • Languages: Python, R, MATLAB
  • ML Methods: Gaussian Processes, Causal ML, Bayesian Methods
  • Cloud: Azure (TotalEnergies infrastructure)
  • Research Tools: Academic publishing, scientific communication

Impact

  • Patent filing for novel prediction methodology
  • 3 scientific publications (ICML 2023, CSDA journal, PhD thesis)
  • 80% confidence in production predictions enabling better operational decisions
  • Presented research to Digital Factory stakeholders for real-world deployment
  • Advanced state-of-the-art in causal inference for continuous treatments

Publications

  1. Acharki, N., Lugo, R., Bertoncello, A., & Garnier, J. (2023). Comparison of meta-learners for estimating multi-valued treatment heterogeneous effects. ICML 2023.

  2. Acharki, N., Bertoncello, A., & Garnier, J. (2023). Robust Prediction Interval estimation for Gaussian Processes by Cross-Validation method. Computational Statistics & Data Analysis, 178:107597.

  3. Acharki, N. (2022). Statistical learning and causal inference for energy production. PhD Thesis, École Polytechnique.

Key Learnings

  1. Academic Rigor + Industrial Impact: Bridging theoretical research with practical applications drives innovation

  2. Uncertainty Matters: Decision-makers need reliable uncertainty estimates, not just predictions

  3. Causal Thinking: Answering “what if” questions requires causal methods, not just correlation

Skills Demonstrated

  • Advanced statistical modeling (Gaussian processes, Bayesian methods)
  • Causal inference and treatment effect estimation
  • Research design and scientific publication
  • Uncertainty quantification
  • Stakeholder communication in industrial R&D
  • PhD-level domain expertise