Fast detection of soil nitrate has an important significance for variable rate fertilization.As NO-3ion selective electrode is easily affected by chloride ion and temperature interference in its practical application for soil nutrient test,a three-layer artificial neural network model optimized by response surface methodology(RSM)was designed to reduce the interference of Cl-on NO-3ion selectivity electrode(ISE)in soil nitrate detection.The output of the model was NO-3-N concentration and three input parameters were temperature,response potential of ISE-NO-3and of ISE-Cl-.A multi-layer feed-forward(MLFF)network with one hidden layer and using gradient descent with momentum(GDM)as learning algorithm was used to develop the error correct model.By response surface methodology,a multivariate quadratic equation was developed to quantitatively describe the relationship between mean absolute error(MAE)and topological parameters of the artificial neuron network(ANN)model,then the optimum number of hidden neurons,momentum coefficient,training epoch,step size,and training runs were found.In range of 10 to 40℃,the best ANN model can correct interference of Cl-within 250 mg/kg while the primary ion concentration ranging from 5 to 250 mg/kg.For practical soil nutrient detection,the MAE could reach 8.6 mg/kg and relative standard deviation was lower than 6.5%compared with16%of no model correction.
