Detection of oil pipelines’ heat loss via machine learning methods

Abstract

In cold climates, when transporting oil with a high paraffin content, oil is heated up to 50 °С. The large surface area, insulation quality and installation conditions also influence heat loss from the main pipeline. Heat losses outcome in warming of the soil and changes in the spatial position of buried oil pipelines. As a result, pipeline operators construct geotechnical networks to measure soil temperature at different depths and to determine spatial changes along the route. Despite the large amount of temperature readings, pipeline operators do not employ sufficient machine learning techniques to detect heat losses. The paper offers an approach that includes generating multidimensional dataset, excluding omissions and outliers, grouping the data by wells, balancing the number of measurements in the groups, performing an upward sampling, training the auto-regressive additive model and separating the trend component from the seasonal one.