TIME SERIES PRODUCTION FORECASTING AND UNCERTAINTY QUANTIFICATION USING PROBABILISTIC DECLINE CURVE ANALYSIS AND MACHINE LEARNING TECHNIQUES - FOR UNCONVENTIONAL RESERVOIRS

Aug 5, 2026

Upload by Violet

Production forecasting and uncertainty quantification remain difficult tasks, especially when combining decline-curve analysis (DCA) models with probabilistic methods or deep-learning approaches. This thesis applies three probabilistic algorithms—Approximate Bayesian Computation, conventional bootstrap, and modified bootstrap—integrated with deterministic DCA models (Arps, Duong, and logarithmic growth analysis). It also employs LSTM and Transformer neural networks, with LSTM hyperparameters tuned using Bayesian optimization. Experiments on 400 Marcellus shale gas wells compare P10/P50/P90 uncertainty bands and MAPE across multiple hind-cast lengths.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages58

Reading Time2 h 26 min

AccessFree to download

Abstract

Acknowledgments

Table of Contents

List of Figure

List of Tables

Nomenclature

Chapter 1: Introduction

Chapter 2: Methodology

Collection and cleaning of the data and Marcellus shale overview

Bayesian Method—approximate Bayesian computation (ABC)

Frequentist method—conventional Bootstrap (CBM)

Frequentist method—modified Bootstrap (MBM)

Machine learning method—Long-short-term memory (LSTM)

TIME SERIES PRODUCTION FORECASTING AND UNCERTAINTY QUANTIFICATION USINGPROBABILISTIC DECLINE CURVE ANALYSIS AND MACHINE LEARNING TECHNIQUES FORUNCONVENTIONAL RESERVOIRS byKapil Hari Sachdev B.Tech,Dr.Vishwanath Karad MIT World Peace University,2021 A Thesis submitted in Partial Fulfillment of the Requirementsfor the Degree of Master of Science inPetroleum Engineering University of Alaska FairbanksAugust 2024 APPROVED: Obadare Awoleke,Committee ChairArghya Das,Committee Co-ChairScott Goddard,Committee MemberAbhijit Dandekar,Chair Department of Petroleum Engineering William Schnabel,Dean College of Engineering and Mines Richard Collins,DirectorGraduate School C Copyright by Kapil SachdevAll Rights Reserved Dedication I want to dedicate this thesis work to the Supreme Personality of Godhead,Lord Shri Krishna,for his mercy and strength. Abstract Researchers have long been engaged in the challenging task of production forecasting anduncertainty quantification,often by combining Decline Curve Analysis(DCA)models withprobabilistic algorithms.Also,deep-learning approaches have been explored for productionforecasting.In this work,we present the application of three probabilistic algorithms combinedwith various deterministic DCA models.We have also applied two machine learning(deeplearning)algorithms for production forecasting and uncertainty quantification.Finally,we havecompared the results of ML algorithms with the probabilistic algorithms based on the obtainedMAPE values to conclude whether probabilistic or ML algorithms are better for productionforecasting. Our analysis commences with the utilization of three probabilistic algorithms,namelyApproximate Bayesian Computation (ABC),Conventional Bootstrap(CBM),and ModifiedBootstrap methods(MBM).Each algorithm was integrated with three deterministic DCA models:Arps,Duong,and Logarithmic Growth Analysis (LGA).We then harnessed the power of machinelearning(ML)algorithms called long-short-term memory(LSTM)and Transformer neuralnetworks.The hyperparameters for the LSTM algorithm were chosen using BayesianOptimization.We conducted a comprehensive study on 400 gas wells from the Marcellusunconventional shale basin,evaluating LSTM,Transformer,and each probabilistic-DCAcombination.Our hindcasting,which employed 12,24,36,48,and 60 months of historicalproduction data(hind-casts)to forecast up to 96 months,yielded forecasts that are essentially10th,50th,and 90th percentiles,providing the P10,P50,and P90 estimates,respectively.Theseestimates,which we refer to as uncertainty bands,vividly demonstrate uncertainty quantificationas we progress from 12-60 months of hind-casts.In simple words,as we increase the use ofhistorical data from 12 to 60 months,the P10 and P90 bands tighten,depicting that uncertaintydecreases in the forecasts.Furthermore,we present the Mean Absolute Percentage Error(MAPE)for each probabilistic algorithm-DCA combination and LSTM algorithm for each hind-cast lengthfor comprehensive comparison and conclusive insights. Our findings are helpful and should inspire confidence in the potential of ML algorithms forproduction forecasting and uncertainty quantification.We demonstrate the superiorperformance of both MLalgorithms (LSTM and Transformers),particularly for 12 to 36 months ofhind-cast for MAPE values and uncertainty bands.The compression of uncertainty bands withincreasing hind-cast lengths indicates a decrease in uncertainty as production history increases;a promising result.Furthermore,the MAPE value decreases as we extend the hind-cast periodfrom 12 to 60 months,suggesting improved accuracy with longer hind-casts. The uniqueness of this work is in the comparison of the ML algorithms(Transformers and LSTM)with the probabilistic DCA approach as discussed above.Also,the proportionality scaling function(PSF)that is introduced in this work allows the analyst to be able to capture the uncertaintyassociated with machine learning forecasts in a

FAQs

What probabilistic algorithms are used for uncertainty quantification in this thesis?

The thesis uses Approximate Bayesian Computation (ABC), Conventional Bootstrap (CBM), and Modified Bootstrap (MBM). Each method is paired with deterministic DCA models to produce probabilistic forecasts and uncertainty bands.

Which deterministic decline-curve analysis (DCA) models are combined with the probabilistic algorithms?

The probabilistic methods are integrated with Arps, Duong, and Logarithmic Growth Analysis (LGA). This combination supports production forecasting with quantified uncertainty.

How are uncertainty bands and forecast accuracy evaluated and compared?

Forecasts are produced as P10, P50, and P90 percentiles (P10/P50/P90), forming uncertainty bands across hind-cast lengths. Accuracy is compared using MAPE values for each probabilistic-DCA combination and for the LSTM/Transformer models.

Aug 5, 2026

Upload by Violet