A comparison of data preprocessing strategies for neural network modeling of oil production prediction

This paper presents a comparison of the different data preprocessing strategies for developing neural network models for prediction of oil production rate. Data processing is an important step in developing a neural network application, which could affect model accuracy and results. We considered the following three ways to preprocess monthly oil production data: (1) the sequential approach in which condition-decision records from all the wells in a reservoir are placed sequentially to form a data set, (2) the averaging approach in which a data set is formed by averaging data values from individual wells in a reservoir, and (3) the individual approach in which data for individual wells are used separately to build models tailored for individual wells. Some advantages and disadvantages, as well as results of each approach are discussed.