The understanding of interwell connectivity of oil reservoirs is of great significance for the formulation of reasonable development and adjustment plans and the improvement of water-driven reservoir recovery. Based on dynamic data of injection-production well，an analysis method combining Kalman filter and artificial neural network is established to quantitatively characterize the dynamic interwell connectivity in reservoir. Considering the noise pollution of the injection data and the time-lag effect of the injection signal in the formation propagation process，the Kalman filter algorithm and the nonlinear diffusion filter are used to pre-process the injection-production data，thereby reducing the effect of injection-production data on the machine learning model，and improving the accuracy of connectivity analysis. Based on the pre-processed historical injection and production data，the artificial neural network taking the oil production rates of producers as the response and the water injection rates of injectors as the input is trained and the parameters are optimized，and the interwell communication relationship in the injection and production system is simulated and excavated. Through the parameter sensitivity analysis of the trained model，the degree of interwell connectivity in reservoirs is quantified. The model is applied to analyze the interwell connectivity in four types of reservoirs with representative characteristics such as homogeneity，anisotropy，closed faults，high permeability zones and one real heterogeneous reservoir. The calculation results are highly consistent with the reservoir geological features，indicating that this method has good practicability and reliability. It can be applied to effectively quantify the connectivity of injection and production system.