Abstract: In the actual production of oilfields, the connection of injection and production is a difficult but important issue. It is of great significance for the formulation of oilfield development plans and the description of remaining oil distribution. In this paper, the dynamic data of a reservoir in Dagang Oilfield is used to establish a MLP neural network model based on Bayesian optimization. The Sobol sensitivity analysis method is used to calculate the sensitivity coefficient. The sensitivity coefficient is used to quantitatively evaluate the connectivity of injection and production. The validity and reliability of the method are verified by comparison with the tracer interpretation results. The research shows that the established Bayesian optimization-based MLP neural network model achieves a good fitting effect, and the Sobol sensitivity coefficient can effectively evaluate the connection of injection and production. The result is consistent with the actual situation of the reservoir.

Key words: injection-production connectivity, Bayesian optimization, MLP neural network, Sobol sensitivity analysis