Units
1. Introduction to Artificial Intelligence and Verification Discipline in Petroleum and Natural Gas Engineering 2. Drilling Data, Well Logs and Petrophysical Interpretation 3. Seismic Data Analysis, Feature Extraction and Structural Interpretation 4. Reservoir Characterization, Facies Modeling and Geostatistics 5. Production Forecast, Drawdown Curve Analysis and Reserve Estimation 6. Drilling and Production Optimization 7. Predictive Maintenance and Equipment Reliability 8. HSE, Process Safety and Operational Risk 9. Pipeline Monitoring, Leak Detection and Integrity Management 10. Data Analysis with Python: Oil-Gas Workflows 11. Regulatory, Ethics, Data Privacy and Holistic Verification Governance
Unit 5 / 11

Production Forecast, Drawdown Curve Analysis and Reserve Estimation

Gains:

  • Ability to explain production forecasting and EUR forecasting with drawdown curve analysis (DCA), Arps models and machine learning
  • Ability to position AI in production data cleaning, drawdown regime selection and reserve classification (1P/2P/3P)
  • Ability to verify production forecast outputs with physical drawdown limits, material balance and reporting standards

When a well is put into production, it gives the highest flow rate on the first day and decreases over time; because the reservoir pressure decreases and the fluid is depleted. Measuring, modeling and extending this decline into the future is the heart of the oil-gas economy. Decline Curve Analysis (DCA) is to estimate future production and total recoverable volume (EUR - Estimated Ultimate Recovery) by fitting a mathematical drawdown curve to production history. This estimate translates directly into the reserve figure and hence the company value, investment decision and regulatory declarations. In this unit, we use artificial intelligence to clean production data, select a drawdown model and generate scenarios. The unchanging rule: EUR is an estimate, not an accuracy; It is never declared without being tested with physical drawdown limits and material balance, especially by taking the optimistic side.

Let's Get to Know Arps Models

Classical DCA is based on Arps' three drawdown models:

  • Exponential drawdown: The drawdown rate is fixed; It is the most conservative (lowest EUR) model. It is close to late period behavior.
  • Hyperbolic drawdown: The rate of drawdown slows over time; It is controlled by a base b. b=0 corresponds to the exponential, b=1 corresponds to the harmonic. Most wells are hyperbolic in the early to middle period.
  • Harmonic drop: b=1; the most optimistic (highest EUR) model. It is rarely valid for the entire life.

The critical point is prime b. As b increases, the drawdown slows down and the EUR grows. If b>1 is taken and extrapolated to the long term, the decline will slow down unrealistically and the EUR will explode. This is the most common over-optimism mistake in the industry.

Caution: In conventional wells b is usually in the range 0–0.5; In unconventional wells such as shale gas/tight oil, b>1 can be seen in the early period, but it cannot be carried to the long term. A minimum decline limit (terminal decline) is always applied in the late period. If AI extrapolated 30 years with b=1.3, this is a red flag.

Reserve Classification

The production forecast is divided into uncertainty levels when converting into reserves:

  • 1P (Proven): High confidence; with reasonable certainty that it will be produced economically. Approximately P90 (90% probability of at least that much).
  • 2P (Proved + Probable): Medium scenario, around P50.
  • 3P (Proved + Probable + Possible): Optimistic scenario, approximately P10.

This classification is governed by standards such as PRMS (Petroleum Resources Management System) and SEC and is the responsibility of the qualified appraiser. AI can help generate scenarios but cannot “declare” the reserve.

Step by step: DCA workflow

  1. Clear data. Closed periods (well shutdowns), measurement errors, interventions (aciding, recompletion) distort the production curve. Tick ​​these; AI can scan for outliers, but you decide.
  2. Select the drawdown regime. Early transient flow or boundary-dominated flow? DCA is reliable only in the border-dominant period.
  3. Make a model. Fit the Arps parameters (initial flow rate, drawdown rate, b); Keep b within physical range.
  4. Apply terminal drawdown. Switch to minimum drawdown in the late period; avoid infinite hyperbolic extrapolation.
  5. Script it. Produce P90/P50/P10 EUR.
  6. Verify. Test with material balance and the neighbor well analogy.

Three Mini Cases: By the Numbers

Case 1 — base b trap. On one tight oil well, AI fitted b=1.5 to the first 18 months of data and extrapolated to 30 years; EUR increased to 620 thousand barrels. When the engineer applied terminal deduction (8%/year), EUR decreased to 410 thousand barrels. Material balance and field analogy supported around 400 thousand. Blind acceptance would overestimate the reserve by 50%.

Case 2 — Dirty data cleaning. Two long periods of downtime (maintenance) appeared as low values ​​in a well's production curve; The drawdown fitted to the raw data showed a "rapid depletion" that did not exist in reality. AI marked stances; When stoppages were removed and actual production days were normalized, the reduction rate turned out to be realistic (14%/year instead of 22%/year). Cleaning corrected the prediction.

Case 3 — Transient flow failure. An engineer applied DCA to the first 3 months of production; This period was still a temporary flow (the border effect had not started). AI emerged from this early steepness with a very high drawdown rate and low EUR. The analysis made after waiting for the transition to the border-dominant period gave the real decrease. Knowing when DCA was valid was essential.

Weak Prompt / Strong Prompt

Weak prompt:

Fit a drawdown curve to this production data and tell the reserve.[data]

Powerful prompt:

DRAFT drawdown curve analysis on the following (anonymised) monthly production data. Rules:- First mark closed period/stoppages, interventions and outliers; REMOVE these from the dream fabrication and write your justification. - Evaluate whether the data is transient or in boundary-dominated flow; DCA is valid only during the border-dominant period, indicate if not appropriate.- Make up arps; Keep the b exponent in the physical range (conventional 0-0.5). Apply a terminal drawdown limit in the late period. DO NOT do infinite hyperbolic extrapolation.- Give EUR as P90/P50/P10, not single number.- Write how to verify with material balance/adjacent well.Data: [production table]

Four Copiable Templates

1) Production data cleaning:

Mark stoppages, interventions, zero/negative values and outliers in the following production data as a "date | problem | suggestion" table. DO NOT CORRECT the data, just mark it. Table: [production]

2) Drawdown regime diagnosis:

From the following production and pressure data (if available), evaluate whether the flow is transient or boundary-dominated. Specify the period during which DCA will be reliable. Explain the risk of adapting it to the early period. Data: [flow/pressure]

3) Arps fitting and scenario:

Fit exponential and hyperbolic Arps to the clean production data below. keep b exponent within 0-0.5; Apply X% terminal drawdown late. Give three scenarios (P90/P50/P10) EUR and write which assumptions are most effective. Data: [clean production]

4) EUR cross check:

Check this EUR forecast: is the cumulative production + remaining forecast sum consistent with the field's material balance/analogue well spacing? List the overly optimistic/pessimistic signs. EUR and inputs: [values]

Drawdown Model Selection Guide

model

b prime

EUR trend

when

exponential

0

Lowest (conservative)

Late mature well

hyperbolic

0–1

medium

Early-middle period, most wells

harmonic

1

High (optimistic)

Rare, use with caution

Unlimited hyperbolic

>1 long term

Too high (incorrect)

Do not use; apply terminal limit

Common mistakes

  • Long extrapolation with high b. b>Moving 1 to years and exaggerating the EUR; skip terminal drop.
  • Fitting it to dirty data. Removing dreams without clearing postures and interventions.
  • Applying DCA to the transient flow. Making predictions before the border-dominant period begins.
  • Odd number EUR. Not reporting uncertainty (P90/P50/P10).
  • Skipping material balance. Not testing the EUR with independent physical/analog control.

In summary

  • DCA is to estimate future production and EUR by fitting a drawdown curve to the production history.
  • Arps b prime is critical; Long extrapolation with high b dangerously overestimates the EUR, terminal drawdown is essential.
  • AI is helpful in data cleaning and scenario generation; but cannot declare the reserve, the qualified appraiser confirms.
  • Report EUR as a P90/P50/P10 distribution; Apply DCA only during the border-dominant period.
  • Verify the result by material balance and adjacent well analogy.

Application task

Get monthly production data for a (representative) well. Generate a DCA draft with a powerful prompt. Then: (1) mark and remove at least one stop/outlier, (2) check the b exponent of the model and apply terminal drawdown if it is above 1, (3) ask for EUR in three scenarios and roughly supply P50 with cumulative + remainder logic.

checklist

  • [ ] I know the Arps exponential/hyperbolic/harmonic models and the effect of the b prime on the EUR.
  • [ ] I am cleaning the production data for downtime/intervention.
  • [ ] I apply DCA only during the boundary-dominated flow period.
  • I keep the [ ] b prime in the physical range and apply terminal drawdown, giving the EUR P90/P50/P10.
  • [ ] Verifies EUR with material balance and analog well, I understand that reserve declaration is the responsibility of the assessor.
Previous Next