Extracting subsurface temperature gradients from radar probes using machine learning with applications to enhanced oil recovery

Technological advances and depletion of easily extracted oil reserves have led to the development of enhanced oil recovery (EOR) methods that allow significantly more oil to be extracted from a reservoir. An increasingly commonly used technique uses thermal injection [1] . Crude oil is heated underground by various methods to affect viscosity and mobility ratio resulting in yield increases up to 60%. Another EOR method, not considered in this paper, is through gas injection. With thermal methods steam is injected into oil wells to improve oil production rates and enhance oil recovery by decreasing oil viscosity. The implementation of oil recovery by steam injection requires a good knowledge of the subsurface thermal conditions. Temperature observation wells (TOW) are used to measure subsurface temperature profiles which can be used to define the vertical thickness of the steam flood, area sweep, rate of heat flow and heat losses to surrounding rock layers. When trying to enhance oil recovery from heavy oilfields, operators may take temperature measurements once every 90 or 180 days for each TOW location. The cost of conducting this invasive, down hole thermometer measurement is around $5,000 per well, apart from the upfront drilling cost. During measurement the wells are not in production and there is additional cost associated with this downtime.