The relative permeability curve is an important basic datum in oil and gas field development research. The traditional experimental method is expensive and time-consuming，and the relative permeability curve obtained by a few test samples is difficult to represent the characteristics of the whole reservoir. The results from the empirical formula method have low precision and large error. To obtain the relative permeability curves of the water-flooding reservoir in real time and accurately，this paper utilizes machine learning algorithms for simulation and prediction. The simulation sample set of the water-flooding reservoir relative permeability curves is constructed by analyzing the sensitivity of logging parameters and integrating the data of the relative permeability curves. On this basis，the machine learning algorithm is selected to optimize the geological factor constraint and the curve endpoint constraint，and the intelligent visualization generation of the relative permeability curves is realized. The results show that this method can realize the prediction of the relative permeability curves for each well and each section，and the prediction accuracy is more than 90%. It can accurately reflect the flow characteristics of reservoir and the variation law of reservoir permeability，possessing high practical application value and a good prospect of popularization and application.