Neural network-based irrigation control for precision agriculture

In the present work, a design of an automatic irrigation neuro-controller for precision agriculture is presented. The irrigation neuro-controller regulates the level of moisture in agricultural soils, specifically in the root zone, using an on-off control-type that opens and closes the valves of the irrigation system (IS). The changes in the moisture levels in the roots area can be modeled as a non-linear differential function depending mainly on the amount of water supplied by the IS, the crop consumption, and the soil characteristics. This dynamic model is identified by a neural network (NN). After the NN is trained, it is used as a prediction model within the control algorithm, which determines the irrigation time necessary to take the moisture level up to a user desired level. At the same time, the NN is re-trained in order to get a new and improved model of the moisture's soils, giving to the IS the capability of adapt to the changing soil characteristics and water crop needs. In this work, it is also presented the main advantages of using this irrigation closed-loop adaptive controller instead of traditional systems that operates to open-loop, such as timed irrigation control.