Deeply understanding the distribution of the flow field in water flooding reservoirs is crucial for improving oil recovery factor and further development of remaining oil. However, while traditional numerical simulation method has been widely applied, its complexity and computational intensity hinder timely adjustments to development strategies. Therefore, there is an urgent need for new technologies to obtain precise and rapid predictions of reservoir flow field characterization parameters, thus promoting efficient oil reservoir development. In this research, through the design of an interface between the numerical simulator and the deep learning platform, an automated process has been achieved from the establishment of numerical simulation models to the construction of sample libraries. Then, based on a deep fully convolutional encoding–decoding neural network, a rapid prediction model for characterizing flow field parameters in water flooding reservoirs has been established. This method not only considers geological and development parameters but also take time series into account. It has demonstrated excellent performance in conceptual models and actual oil reservoir models, with prediction decision coefficients exceeding 0.90. By analyzing the effect of model dimension numbers, the flow field influencing factor numbers, and the number of samples on prediction performance, it has been found that increasing the number of samples can improve prediction accuracy. This method effectively solves the problem of predicting flow field characterization parameters under the influence of complex factors, significantly improves prediction efficiency, and can obtain the prediction results in just one second, which provides strong technical support for efficient oilfield development.