Neural compensator for PI soil moisture control

Abstract

The spatial and temporal variability of a cultivated soil, with technified irrigation systems, requires adaptive control systems to the varying conditions of the water–soil–crop intersystem. Therefore, an adaptive control based on a Radial Basis Function Neural Network (RBF-NN) is proposed in this paper. A static Proportional-Integral (PI) controller was tuned without modifying its parameters by adding a compensation based on RBF-NNs. In this way, the dynamic variation is approximated in real time by means of a RBF-NN. The controller is tested in simulation from a model of water distribution in the soil with extraction by a crop. The results obtained with this method are compared with a traditional Proportional-Integral-Derivative (PID) controller. The comparisons are made taking into account compromise between the amount of water applied and irrigation frequency to keep soil moisture values within the allowed limits. Water savings of 20% and a reduced valve activations 2 times less than the traditional PID were achieved. Finally, the behavior of the controller in the event of disturbances was evaluated, verifying the rejection it produces in the face of these eventualities.