One of the crucial components of any company successful development in the oil industry is the high-quality processing and analysis of a large amount of data for subsequent solving the forecasting and scheduling tasks of the oil, liquid, natural and co-produced wellhead gas production. Under the conditions of rapidly developing IT sphere, the use of machine learning methods is a relevant and a promising direction. However, most of the emerging engineering challenges cannot be solved efficiently by using either only machine learning algorithms or only physical and mathematical models.
Solving the above-mentioned tasks using only one of the approaches is either more labour-intensive (the description of all processes running in the system like in a complete physical/mathematical model), or allows for the possibility of non-physical solutions and high error values (when only machine learning approach is used) in comparison with the combined physical/mathematical and machine learning models.
The proposed hybrid approach allows to eliminate the uncertainties inherent in physical and mathematical models that are difficult to describe analytically by the application of machine learning methods to refine the results.