Hydrocarbon Production Forecasting via Dynamic Rescaling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional Decline Curve Analysis (DCA) for hydrocarbon production forecasting is inadequate due to its inability to capture reservoir signals not modeled, neglect of production history from other wellbores, and lack of utilization of wellbore properties, leading to inaccurate and inconsistent forecasts with high bias and standard deviation.
Innovation Solution
A method involving Dynamic Production Rescaling (DPR) transformation and back-transformation algorithms, which normalize production data across training and target wellbores to enable accurate forecasting by a machine learning algorithm, considering geographic, geologic, and completion data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional DCA models are used for production forecasting, then the forecasting process is simple and widely applicable, but the forecast accuracy is insufficient due to inability to capture reservoir signals not modeled
Solution Approach 1:
The production data is segmented into multiple time periods (first period of time for training, second period of time for forecasting) to enable different processing approaches for different temporal ranges, improving accuracy without requiring a completely complex model
Solution Approach 2:
The DCA model parameters are made dynamic by allowing them to vary across different time periods and wellbores, rather than using fixed parameters, which captures reservoir signal variations while maintaining the familiar DCA framework
2Measurement precision
If traditional DCA is applied to individual wellbores, then the analysis is straightforward, but valuable production history from other wellbores is not considered
Solution Approach 1:
Production data from multiple wellbores is merged into a unified training dataset, allowing the model to learn from collective production patterns across the field while maintaining the ability to make wellbore-specific forecasts
Solution Approach 2:
The forecasting system serves multiple functions: it analyzes individual wellbore production, learns from collective field patterns, and provides forecasts for both historical analysis and future planning, maximizing information utilization
3Measurement precision
If multivariate regression and machine-learning methods are used, then more data can be processed, but the methods are sub-optimal in correlating noisy historical production data with future production
Solution Approach 1:
The production data undergoes preliminary transformation and normalization before being input to the forecasting model, which prepares the data by removing noise and standardizing formats, making the subsequent correlation and forecasting more effective
Solution Approach 2:
A data transformation layer acts as an intermediary between the raw noisy production data and the forecasting model, preprocessing the data to enhance signal quality and improve the model's ability to detect meaningful correlations
4Measurement precision
If recent machine-learning methods are applied, then more variables can be considered, but the results are inconsistent and inaccurate with large bias and standard deviation
Solution Approach 1:
The model transforms and standardizes key parameters such as production rates and time variables before processing, which reduces variability and bias in the results, leading to more consistent and reliable forecasts across different wellbores and time periods
Data Source
AI summary
Embodiments of forecasting hydrocarbon production are provided. One embodiment comprises transforming production data of a plurality of training wellbores and production data of at least one target wellbore such that the production data for all training wellbores and target wellbores are equivalent at the end of a first period of time; generating a transformed production forecast for each target wellbore at a target forecast time responsive to the transformed production data; and back transforming the transformed production forecast for each target wellbore to remove the equivalency. The back transformed production forecast for each target wellbore is the final production forecast for each target wellbore.