Prediction of NO3, K, Ca, and Mg Ions in Hydroponic Solutions using Neural Network Model with an Array of Ion-Selective Electrodes

The measurement of individual nutrient concentrations is crucial in closed hydroponics because the correction of each deficient nutrient can allow both improved efficiency of fertilizer use and increased time of use of the nutrient solution. In this study, back propagation artificial neural network (ANN) algorithm combined with a two-point normalization method was employed to predict NO3, K, Ca, and Mg ions. Data from a sensor array of electrical conductivity (EC), and ion selective electrodes (NO3, K, and Ca) was used as inputs of ANN model. For the training and validation of the model, a set of samples with a background referring the Hoagland‘s solution were prepared by a fractional factorial design with three levels of concentrations for four ions. To compensate the drifts during the measurements, a two-point normalization method was applied prior to each sample measurement by an automated test stand. The prediction model for Mg ions showed a low coefficient of determination (R2=0.37) from the training. However, the ANN models for NO3, K, and Ca ions showed a high modelling capacity with high coefficients of determination (R2>0.9). In application of the models to hydroponic samples from lettuce and paprika growing beds, mean absolute relative errors of the ANN models were 3.1, 16.7, and 4.1% for NO3, K, and Ca ions, respectively.