There are a variety of ions in soil suspension, so when using ion selective electrode to detect soil nitrate-nitrogen, it is particularly vulnerable to the interference of other ions, which will greatly affect the accuracy of detection. In order to improve the accuracy of monitoring soil nitrate concentration by ion selective electrode, it is necessary to establish a model to predict the concentration. The prediction performances of Nernst, SAM and BP-ANN models for soil nitrate concentration detection were discussed. The standard solution test was carried out to predict the prediction performance of the three models. Combined with the field corn monitoring experiment and potted pink crown tomato monitoring experiment to verify the sample experimental results, the soil nitrate concentration was taken as the output of the three models, and the effect of optical method on the determination of soil nitrate concentration was compared. The experimental results showed that the prediction results of the three models were in good agreement with the true values of the samples. Among them, the SAM model was the most accurate, and its correlation coefficients were not less than 0.9. The variation ranges of MAE, MRE and RMSE were 2.03~5.08mg/L, 0.64%~8.79% and 2.21~5.49mg/L, respectively. SAM concentration prediction model had the characteristics of high precision and good anti-interference. It can run on a relatively simple platform and had a wider range of application. In addition, when combined with the fluid control system, it can ensure the detection accuracy and improve the detection efficiency. It had a certain reference value for in-situ detection of soil nitrate nitrogen based on ISE.