Knowing the injection profile of each injector in a reservoir is of major importance for analyzing residual oil distribution underground. However, in practice, it is impossible to acquire real-time injection profiles for each well when they are needed.

In this paper, a two-step learning fuzzy system was introduced to predict injection profiles for wells with outdated data or without data by analyzing those injectors whose injection profiles are well characterized. Five pivotal parameters were selected to be fed into the subtractive clustering algorithm and ANFIS (adaptive-network-based fuzzy inference system) to model their non-linear relationships with the injectivity of each producing layer. Experiments in comparing the two-step method and the backpropagate feed-forward (BPFF) neural network showed that the two-step method can automatically generate a more simply structured fuzzy inference system (FIS) and can simulate complicated numerical nonlinear relationships with higher accuracy.

This method was used in injection profile prediction in one block of the Daqing oilfield. The average accuracy was shown to be above 80%.

Keywords:
accuracy, producer, identification, node, injection profile, thickness, anfis, machine learning, neural network, upstream oil & gas
Subjects:
Information Management and Systems, Artificial intelligence
Copyright 2004, Society of Petroleum Engineers
You can access this article if you purchase or spend a download.