Experimental model and neural network based electrical conductivity estimation in soilless culture system.

Abstract

The optimum management of nutrient solution in soilless culture needs the accurate control of nutrient solution, especially in recycled soilless culture system. To keep the electrical conductivity (EC) of nutrient solution within the adequate range after application of combined fertilizers, theoretically derived EC prediction methods are required. In this study, the experimental EC prediction equation, an extended form of the Robinson and Stroke's theoretical equation only available for a binary electrolyte, was developed for predicting the EC of the nutrient solution containing many kinds of inorganic compounds. And the multilayer perceptron consisting of three layers with the back propagation learning algorithm was developed for EC prediction. It consists of nine variables in the input layer for the concentrations of seven macro elements, Na+ and Cl, and one variable in the output layer for the EC of nutrient solution. The predicted ECs by experimental model as well as neural networks for the nutrient solution were compared to the measured ones and showed good agreements.