Prediction of spreading processes using a supervised Self-Organizing Map

A novel technique is presented based on self-organizing neural networks for prediction of fertilizer distribution patterns of spreaders as a function of spreader settings and fertilizer properties. The main aim of the presented technique is to predict tendencies in the spreading distribution pattern as a function of machine configurations and physical fertilizer properties. The Self-Organizing Map is used in a novel way to represent input–output relationships between high-dimensional spaces. Other NN methods would be very difficult to use because of the high dimensions of the input and output spaces. In the case of a multilayer perceptron, the global connectivity would lead to a prohibitively large number of free parameters giving rise to learning time problems. The spreading distribution patterns are predicted with a high performance with the proposed technique.