Desenvolvimento de modelos em redes neurais para simulação de precipitação na irrigação por aspersão.

Abstract

The efficiency of a sprinkler irrigation system depends on the performance of the sprinkler head, which is dependent on certain characteristics such as nozzle diameter, operating pressure, wind speed and direction, and layout and spacing in the field. Determining the coefficients of uniformity and the efficiency of sprinkler systems usually depends on field trials requiring time and financial resources. One alternative to reduce time and expense is the use of simulations to estimate the values of water distribution or the coefficients of uniformity and efficiency. In this study, we proposed and tested the application of a multilayer perceptron-type artificial neural network (ANN) to simulate the precipitation of a sprinkler having as input parameters the values of operating pressure, wind speed, wind direction, and sprinkler nozzle diameter. Trials were performed in the field with sprinklers operating in a grid of 16 x 16 collectors with 1.5 meter spacing, with different combinations of nozzles, pressures, and wind conditions. An artificial neural network was trained to simulate and estimate the water distribution values for the sprinkler at conditions tested. The ANN model showed good results in the simulation of precipitation, with the Spearman correlation ( ) between the data obtained in the field trial and the simulated data having values between 0.92 and 0.97 for the ten trials analyzed. For the correlation between the distribution profile with the data simulated by the ANN model and the data obtained in field trials, for the same ten trials, R2 coefficient values of 0.95 to 0.99 were observed. The ANN model shows promise in the simulation of water distribution of sprinklers in sprinkler irrigation systems.