In recent years, technology that harnesses the unlimited potential of microorganisms has become important as a modest but long-lasting technology. In order to maximize the power of microorganisms, it is necessary to control the flow of culture medium to mix them uniformly, light, carbon dioxide. Taylor vortices are considered suitable for agitated culture of plant and animal cells or microorganisms because they are easy to create and are resistant to disturbances, stable, and have little local shear flow. In this study, we constructed a system that can automatically discriminate the flow mode using numerical results of Taylor vortex flow generated between rotating double cylinders as input data by using deep learning. By comparing the loss and accuracy rate of test data for various physical quantities and comparing the accuracy rate and loss of training data, the physical quantities that can efficiently predict the mode development process of the Taylor vortex were shown. The results show that among the various physical quantities, the radius u is the most accurate when comparing the final loss after learning and the accuracy rate, and can efficiently predict mode development process of the Taylor vortex.