Abstract

We present an optimized training and prediction model for Rate of Penetration (ROP) forecasting using on-line artificial neural network (ANN) in real-time. The technique aims to assist decision making on drilling operations by predicting ROP under a given set of drilling conditions. The scenario modeler relies on real time drilling data analysis, and it is capable of handling cumulative information analysis in real time for ROP prediction within the same well, but also can consider drilling data gained from other fields under similar conditions. The real time prediction model has been applied to drilling data coming from a geothermal project of over 4 km depth, located in the Pohang, Republic of Korea. The observed results with respect to data intervals or sections set the basis for further adjustments to the model, and encourages its use in different drilling situations.

1. Introduction

ANN have been applied to a wide variety of field research areas that include computer vision, speech recognition, and petroleum engineering (Bilgesu et al., 1997). Especially, in the aforementioned field, ANN have been used for ROP forecast (Gidh et al., 2012). It has been shown that this technique is dependent on the size and accuracy of the input parameters, and in general the more number of data points, the better the results. Its ability to consider more drilling parameters into the model makes it advantageous (Monazami et al., 2012).

During ANN learning phase, a selected group of input parameters are provided to the model and they serve to train the algorithm. Two basic types of learning modes can be mention, On-line and off-line training, and they distinguish from each other basically on the training cases are managed after training (Shin 2001). In on-line training, the provided input parameters are discarded after being processed, however the weights are updated.

Owing to the accumulative manner the drilling data is generated in the field, this work explores the applicability of an artificial neural network with an on-line training mode for ROP prediction especially in subsequent drilling sections within the same well.

This content is only available via PDF.
You can access this article if you purchase or spend a download.