DEEP ARTIFICIAL NEURAL NETWORKS FOR FORECASTING DEBIT VALUES FOR PRODUCTION WELLS

Abstract

The relevance of the research is caused by the need for supporting the decision-making by specialists of the extracting enterprises of the oil and gas industry at production management. Precise forecasting of production wells debits values allows defining such technological operating modes of a well stock and processing equipment which would allow reaching the prescribed production volume. The existing methods do not always provide the demanded precision level at the forecast of wells debits values that leads to mistakes at the calculation of economic effect at a profitability assessment of production wells and the subsequent deliveries of hydrocarbonic raw materials, and also at taking into account the restrictions imposed by environmental supervision of natural resources use. The main aim of the research is to develop and offer the most effective models of deep artificial neural networks at the forecast of production components values for hydrocarbonic raw materials – oil, gas, fluid (gas condensate), and water. Objects of the research are technological parameters of production wells debits of well-stock of oil, gas, oil and gas, and oil-gas condensate fields. Methods: the methods of the analysis of a large amount of technical data of wells developed according to the concept of «Big Data»; models of deep artificial neural networks; object-oriented programming; methods of an assessment and the statistical analysis of research results of deep artificial neural networks efficiency at the forecast of production wells debit values. Results. The technique of data preparation for wells debits is developed for training and testing of feed-forward deep artificial neural networks. Research is carried out for various architectures for such artificial neural networks at the solution of the forecasting task of oil, gas, fluid (gas condensate), and water debits. The most effective architecture of feed-forward deep artificial neural networks is developed. Such neural networks allow increasing the forecasting accuracy in two and more times in comparison with the accuracy of the forecasting received by a traditional method of extrapolation (moving average).

Keywords

• oil and gas field
• production well
• forecasting of wells debits values
• models of deep artificial neural networks
• research of efficiency of feed-forward deep neural networks