In order to ascertain the change law of the substrate moisture content at different depths, drying method was used to calibrate multiple EC-5 sensors, and the four sensors were placed at four different depths, i.e., 5cm, 10cm, 15cm and 20cm vertically from the dripper. The changes of the vertical substrate water content under different dripper flow and drip irrigation conditions were measured, and a prediction model of substrate water content at different depths was established. The test results showed that the substrate moisture content of the first layer (5cm away from the dripper) was risen first after the drip irrigation started and quickly reached a higher level. After the drip irrigation stopped, the moisture would quickly diffuse to the deeper substrate layer, and its moisture content can be increased to the root system easy to use level (25.3% and above), the rapid water migration time lasted for about 1h. With the decrease of the initial substrate moisture content, under the same dripper flow and irrigation conditions, the degree of water migration in the vertical direction was deeper. The initial water content of the first layer of the substrate, drip irrigation time, prediction time, predicted layer height difference, and dripper flow were used as input, and genetic algorithm optimized BP neural network algorithm and random forest regression algorithm (RFR) were used to establish different depths of water content of the substrate under drip irrigation rate prediction model. The predicted water content of the substrate after drip irrigation in the experiment compared with the actual measured water content of the substrate at different depths, and the error analysis was performed on the prediction results of different prediction depths. The results showed that the prediction accuracy (R2) of the GA-BP prediction model and the RFR prediction model were 0.8664 and 0.9465, respectively, that was, the prediction model established by the RFR algorithm was more accurate, and the closer the prediction depth was to the first layer, the more accurate the prediction result was.