Data Driven Production Forecasting of Unconventional Wells with Apache Spark

ABSTRACT

Real-time decision making, field surveillance, and production optimization improve the performance of existing operations to increase hydrocarbon recovery and reduce emissions. In this regard, the oil and condensate flow metering in offshore gas condensate platforms is always confronted by environmental, economic, and operational challenges resulting in uncertain production management plans. Although production forecasting of unconventional gas condensate systems is more challenging than for conventional wells, it is of great interest to support decisions by knowing the future of the wells as far as possible. The virtual flow metering techniques make it possible to utilize daily production data sets and extract information on how wells and reservoir will respond to different operational conditions. The objective of this study is to embed artificial intelligence algorithms in reservoir uncertainty modeling and present a mechanistically-supported data-driven model applicable for production forecasting of gas condensate wells with higher confidence. The outcome entails a new set of mathematical models, implemented using Apache Spark cluster computing engine with APIs in Python, that enables rigorous and robust optimization of the recovery process, designing and discovering hidden patterns in production data, and extracting reservoir information indirectly in seconds. The observations used to demonstrate the performance of the proposed hybrid model include 1600 well-testing data points together with 420 days of production history of an offshore gas condensate platform. The daily platform production is allocated efficiently to individual wells using a multilayer perceptron neural network model adaptively trained with well-testing and daily production datasets, and supported by the Energy and mass balance equations.

A full version of this paper is available on OnePetro Online.