Artificial neural network–based time-domain interwell tracer testing for ultralow-permeability fractured reservoirs

Abstract Affected by water injection, the interwell parameters of injection-production wells in ultralow-permeability fractured reservoirs are changed in the time domain, and interwell tracer testing can characterize interwell characteristics quantitatively. For an oil field in the middle and late stages of development, the number of interwell tracers is reduced to control costs, meaning that some injection-production well groups cannot be monitored effectively. Given the rapid development of artificial intelligence technology by which the complex implicit relationship of parameters can be obtained through sample learning, it is possible that future interwell parameters can be predicted by combining a large accumulation of data from interwell tracer testing of well groups with historical production data. This work took a shallow, ultralow-permeability, fractured reservoir in the eastern part of the Ordos basin as its object, and based on the interpreted parameters of previous interwell tracer tests, combined with the dynamic production data and static geological parameters, an analysis method of time-domain interwell tracer testing was developed for ultralow-permeability fractured reservoirs using an artificial neural network. Learning from actual field samples and analyzing two schemes of neural network structure, the minimum error was found for the scheme that predicted the next interwell parameter based on the interpreted parameters of the last time the tracer reached the minimum, with a relative error of 1.94%. The proposed method can predict future interwell parameters reliably and can provide a new low-cost monitoring method for the middle and late stages of water injection development in ultralow-permeability fractured reservoirs.