The key factor to control the shape of salt cavern cavity is to control the depth of the interface between solvent and brine, and the accurate monitoring of the interface is the prerequisite for its control. The current two-phase media interface monitoring technology has the disadvantages of poor real-time performance and low precision. A special algorithm for two-phase media monitoring in cavity well was proposed. Distributed optical fiber was used to collect the temperature data in the vertical direction of the cavity, and the original temperature curves were respectively processed by local weighted regression analysis and Kalman filtering. Through the difference calculation of the treated curve, the region with the most obvious temperature difference was identified, and the interface position of the two-phase medium was preliminatively determined. Then, the temperature curve in the initial location area was processed by convolutional smoothing filtering, the weighted rate of change of the convolutional smoothing filter curve was calculated, the maximum value of the temperature change rate was found, and the specific location of the interface of the two-phase medium was finally determined. The algorithm was used to monitor the interface of two phase media under actual conditions. The experimental results show that the error of the measured data obtained by this system is controlled within 0.3 m compared with that obtained by neutron logging method. Compared with the traditional optical fiber monitoring technology, the proposed algorithm has the advantages of simple operation, high measurement accuracy and high reliability.