Daily Production Prediction Model of Oil Wells Using Backpropagation Neural Network

Abstract

Background: Measuring well production is a crucial task in upstream oil and gas operations, where various tests and measurements are standard procedures and integral parts of operational activities. More frequent production measures are required to detect production declines in mature fields. However, existing daily production testing at the Langgak field, Central Sumatra Basin, cannot be routinely and periodically conducted due to several economic and technical challenges. 
Purpose: The objective of this article is to create a model for predicting daily crude oil well production.
Design/methodology/approach: To achieve this goal, the study applies an artificial neural network (ANN) for forecasting daily crude oil well production, utilizing 17,394 daily production records from 26 wells. This sample size is well above recommended thresholds for neural network models, ensuring sufficient data for robust model training and validation. The backpropagation algorithm and the sigmoid function are employed as the learning algorithm to predict daily crude oil well production. 
Findings/Result: The optimal parameters for predicting daily crude oil well production were 20 hidden nodes and a learning rate of 0.05, converging at 481 epochs with a training time of 13 seconds.
Conclusion: Model performance was indicated by high correlation coefficients (R) across training, validation, and testing phases, along with a low Mean Squared Error (MSE). The resulting regression equation, Output = 1 × Target + 0.00024, confirms a near-perfect alignment with the target function.
Originality/Value (State of the art): Although this study employs the standard backpropagation neural network (BPNN) architecture an established method in oil production forecasting it contributes original value by rigorously applying 10-fold cross-validation on an 80:20 train-test split of the Langgak field dataset, thereby enhancing model reliability and offering validated insights for forecasting in mature oil fields; a foundation upon which future research can build using hybrid or more advanced neural architectures shown to yield superior accuracy.

Keywords: artificial neural network, backpropagation, oil production forecasting, machine learning, yield prediction